How does dollarization affect real volatility and country risk

Universidad Nacional de La Plata Quintas Jornadas de Economía Monetaria e Internacional La Plata, 11 y 12 de mayo de 2000 How does dollarization affect real volatility and country risk? Jorge Eduardo Carrera (CACES-UBA y UNLP), Mariano Féliz (CACES-UBA, PIETTE-CONICET y UNLP) y Demian Tupac Panigo (CACES-UBA, PIETTE-CONICET y UNLP) How does dollarization affect real volatility and country risk?* 1 Jorge Eduardo Carrera (CACES-UBA, UNLP) Mariano Féliz (CACES-UBA, PIETTE-CONICET, UNLP) Demian Tupac Panigo (CACES-UBA, PIETTE-CONICET, UNLP) Version: January/2000 Abstract This study gives a non-traditional framework for the evaluation of the convenience of an asymmetric monetary association (such as dollarization), from the point of view of the country that gives up its monetary sovereignty. In the analytical part we discuss the relationship between nominal volatility, real volatility and country risk. Given the social loss function of the policymaker, we determine the necessary conditions for dollarization to improve social welfare. With this in mind, we concentrate in the analysis of two main aspects: 1) the degree of synchronization existing between the cycle of the leader and associated country, and 2) the effect and relative importance of the different channels (the trade and financial channels) that transmit the shocks from the central country (the United States). In the empirical part we perform an application of our analytical framework to the case of Argentina. To estimate the synchronization of the business cycles we use the coefficient of cyclical correlation, calculated for four different methodologies of de-trending. The effect and relative importance of the financial channel and the trade channel were extracted from the impulse-response functions and variance decompositions of a Vector Error Correction Model (VECM). We analyze the stability of the results altering the order of the variables, re-estimating the model with rolling sub-samples and modifying the deterministic component in the error correction mechanism. As a general result dollarization in Argentina would not only reduce the risk of devaluation but also reduce the real volatility of the economy and so the country risk. For that reason, from the financial point of view the advantages of dollarization will depend on how much society values the alternative of keeping open the possibility of adjusting to extraordinary shocks with the exchange rate parity. Keywords: monetary union, dollarization, VECM, risk, volatility, transmission channels, JEL codes: C5, F3 * The views expressed in this paper do not necessarily represent those of the institutions to which the authors belong. As usual, mistakes and omissions are the authors exclusive responsibility. 1 jcarrera@isis.unlp.edu.ar How does dollarization affect real volatility and country risk?* 1 Jorge Eduardo Carrera (CACES-UBA, UNLP) Mariano Féliz (CACES-UBA, PIETTE-CONICET, UNLP) Demian Tupac Panigo (CACES-UBA, PIETTE-CONICET, UNLP) Version: January/2000 Contents 1. INTRODUCTION .................................................................................................................................................................................1 2. TRADITIONAL REQUISITES FOR A MONETARY UNION ......................................................................................................................2 3. COSTS AND BENEFITS OF DOLLARIZATION .......................................................................................................................................2 4. DOLLARIZATION , NOMINAL VOLATILITY AND REAL VOLATILITY ...................................................................................................4 5. VOLATILITY AND COUNTRY RISK .....................................................................................................................................................5 6. HOW DOES THE UNITED S TATES’ CYCLE TRANSMIT IN THE CONTEXT OF AN ASYMMETRIC MONETARY UNION ? .........................6 7. THE PROBLEM OF THE POLICYMAKER CONFRONTED WITH THE ALTERNATIVE OF DOLLARIZATION .............................................8 8. DOLLARIZATION , REAL VOLATILITY AND COUNTRY RISK. AN APPLICATION TO THE CASE OF ARGENTINA ...............................12 8.1. 8.2. 9. EFFECT AND RELATIVE IMPORTANCE OF THE DIFFERENT CHANNELS OF TRANSMISSION OF THE BUSINESS CYCLE. A VEC MODEL APPROAC H ..............................................................................................................................................................................13 CORRELATION BETWEEN THE CYCLES ......................................................................................................................................20 CONCLUSIONS................................................................................................................................................................................22 10. REFERENCES .............................................................................................................................................................................24 11. APPENDIX ..................................................................................................................................................................................28 11.1. 11.2. 11.3. 11.4. 11.5. 11.6. 11.7. THE DATA: VARIABLES, SOURCE, FREQUENCY AND SAMPLE SIZE ...............................................................................................28 RELATIONSHIP BETWEEN REAL VOLATILITY AND COUNTRY RISK (ARGENTINA) .............................................................................29 RELATIONSHIP BETWEEN THE VARIABLES OF THE HYPOTHESIS IN EQUATIONS (2), (3), AND (5) ....................................................30 UNIT ROOT TESTS ..................................................................................................................................................................31 TEST TO DETERMINE THE OPTIMAL LAG LENGTH FOR THE VEC MODEL .......................................................................................32 UNIT ROOT TEST FOR THE RESIDUAL FROM ERROR CORRECTION MECHANISM. ...........................................................................33 SENSITIVITY ANALYSIS RESULTS. .............................................................................................................................................34 Keywords: dollarization, VECM, risk, volatility JEL classification codes: C5, F3 * The views expressed in this paper do not necessarily represent those of the institutions to which the authors belong. As usual, mistakes and omissions are the authors exclusive responsibility. 1 jcarrera@isis.unlp.edu.ar How does dollarization affect real volatility and country risk?* 1 Jorge Eduardo Carrera (CACES-UBA, UNLP) Mariano Féliz (CACES-UBA, PIETTE-CONICET, UNLP) Demian Tupac Panigo (CACES-UBA, PIETTE-CONICET, UNLP) Version: January/2000 1. Introduction Dollarization is a key issue in the economic policy agenda of several emerging countries. After the crisis of the nineties, the discussion as regards the virtues of the different exchange rate regimes has reappeared. 2 Nowadays, the alternative of dollarizing an economy has become extremely relevant, in academic as well as in political spheres. In Latin America the proposal has been discussed in general terms by a number of economists and even at an official level it has been presented and discussed by the President of Argentina's 3 Central Bank (BCRA) . At the same time, the Federal Reserve and a United States Senate committee have evaluated Argentina's proposal of a Treaty of Monetary Association. While some aspects of the discussion on the costs and benefits of dollarization relate to the contrast between flexible and fixed exchange rate 4 regimes, a deeper analysis requires advancing in other aspects of a process of dollarization . Most of the studies on monetary unions have centered on the transmission of shocks through trade flows. On the contrary, the main objective of this paper is to concentrate on the financial aspects of a monetary association given the fact that, contrary to the experience of the 60s (when the optimal currency area theory was born), the freedom of capital movements of today has increased the importance of financial flows in the determination of the costs and benefits of dollarization. For that reason we focus our analysis on the relationship between dollarization and volatility. We show how recent literature has attributed importance mainly to nominal volatility, like the case of excessive nominal volatility caused by central bank's discretionary policy (political shocks), or the nominal volatility needed for nominal contracts to work as a mechanism of hedging against economic shocks. We introduce the effects of real volatility and its direct influence on country risk. While dollarization directly reduces the risk of devaluation, its effects on the country risk are ambiguous. To make a precise statement with respect to the final effect of dollarization in financial terms, we must find out the sign and magnitude of its effect on the real volatility of the economy and thus on country risk. In the analytical part of the paper we discuss the concepts of real and nominal volatility, the behavior of the channels of transmission of external shocks and the relationship between real volatility and country risk. We define the objective function of the policymaker and establish the necessary conditions for dollarization to reduce the aggregate risk (that is, the sum of devaluation risk plus country risk). In the empirical section we apply this analytical framework to the case of Argentina. To estimate the association of the business cycles we use coefficients of cyclical correlation calculated from four different detrending methodologies. The effect and relative magnitude of the financial and trade channels were extracted from the impulse-response functions and variance decompositions of a Vector Error Correction Model (VECM). We analyze the stability of the results altering the order of the variables, re-estimating the model with rolling sub-samples and changing the deterministic component in the error correction mechanism. * The views expressed in this paper do not necessarily represent those of the institutions to which the authors belong. As usual, mistakes and omissions are the authors exclusive responsibility. 1 jcarrera@isis.unlp.edu.ar 2 In general terms, dollarization can be understood as the resignation that a country makes of its monetary sovereignty to introduce the US dollar as its domestic currency. 3 Pou, P.(1999) "Más dolarización para profundizar la convertibilidad", Clarín 05/26/1999. Other Latin American countries, such as Ecuador, are discussion this alternative. See Roubistein (1999) for an approach favorable to dollarization and Posen (2000) for an analysis of the position of the United States on the issue. 4 The analytical part of this paper is also useful for the discussion of the case of "euroization" that is of interest in several European countries. On this issue, see Posen (2000). 1 Finally, in the conclusions we combine the deductions of the analytical section with the results of the empirical part. 2. Traditional requisites for a monetary union The dollarization of an economy can be understood as the conformation of a monetary union between a country that substitutes the US dollar for its domestic currency. The important characteristic of this type of union is that it is asymmetric, in the sense that there is a big country which acts as leader (the United States) 5 while the rest of the countries (those opting for dollarization) act as followers . To analyze the effects of a process of dollarization it is useful to review the recommendations of the theory of optimum currency areas (OCA). If the countries fulfil the conditions for an OCA then they are able to optimally coordinate their economic policies. This theory initiated in the sixties with the works of Mundel 6 (1961), McKinnon (1963) and Kenen (1969) . The traditional elements required to evaluate the optimality of a "monetary association" between two countries are: 1. The degree of similarity in their economic structures (Kenen criterion): the greater the degree of similarity in the structures, the greater the effects of external common shocks and the higher (and positive) the correlation between the business cycles. 2. The level of integration of the economies as measured by the volume of total trade between the countries (McKinnon criterion): the more integrated two nations are, the greater the transmission of shocks between them and thus, the more correlated their business cycles will be. 3. The existence of trade not based on comparative advantages: a high degree of intra-industrial trade will contribute to the equalization of the economic structures making shocks more similar. 4. Price and wage flexibility: the more rigid the nominal variables are, the most useful the exchange rate 7 policy becomes as an instrument for changing relative prices . 5. Mobility of factors of production such as labor and capital across countries or regions (Mundel criterion): the greater the mobility of factors in response to asymmetrical shocks, the greater the compensatory flows of factors will be (including migrations). 6. The existence of inter-jurisdictional fiscal transfers: this allows for countries to compensate for asymmetrical shocks with an instrument different from that of changing the bilateral exchange rate. The common argument in the traditional framework is that if shocks are similar (that is, if we find that the correlation of business cycles is positive, with similar intensity and duration), then using exchange rate policy (between partners) is not effective to compensate for shocks. 3. Costs and benefits of dollarization To evaluate the specific impact of dollarization it is convenient to review synthetically the traditional view of the costs and benefits of a monetary union. The benefits can be grouped, according to Fenton and Murray (1993), in four types: reducing transaction costs, reducing uncertainty, improving credibility and anti-inflationary discipline and improving the behavior of 5 The level of asymmetry is a characteristic that differentiates dollarization from the European Monetary Union (see Cohen and Wyploz, 1990, or De Grawe, 1992). While in Europe the issue refers to the degree of marginal influence of each country in the decisions of a supra-national institution, in the case of dollarization it seems difficult to think about the incorporation of the countries, abandoning their domestic currencies, into the Board of the Federal Reserve Bank of the United States. The are several papers that study the problem of monetary unions in the framework of game theory under a symmetric set up. However, only few do so in an asymmetrical context. Amongst these we find the work of Canzoneri, Henderson and Sweeney (1987) for a game between Europe and the United States, Cooper (1991) for an application of a Stackelberg solution and Carrera (1995) for the specific case of an asymmetric monetary union within a leader-follower framework. 6 In the nineties there has been a revalorization of this theory (for an analysis with recent developments see Masson and Taylor, 1991). In the theoretical aspects of the discussion Cassella (1993) gives microfundations to the OCA. Ghosh and Wolf (1994) have established a genetic approach to define its optimality and Mélitz (1991) has suggested an important theoretical reformulation. As regards the empirical aspects of the discussion Bini Smaghi and Vori (1993) study the European case and Chamie, DeSerres and Lalonde (1994), Bayoumi and Einchengreen (1992) and Rogoff (1991) study the case of NAFTA. 7 For shock asymmetry to be a problem, a certain price and wages inflexibility (especially downwards) is required. If domestic prices have great flexibility, countries can do without other auxiliary policies to absorb shocks. If prices have great downward flexibility (in degree and velocity), when an economy is hit by a negative shock there is no need to use devaluation as a substitute; if prices are extremely flexible upwards, devaluation is useless. In other words, to justify the independence of exchange rate policy we need to find that the variation of the nominal exchange rate will generate a change in the real exchange rate (Bootle, 1995). 2 the monetary system. Amongst the costs we find the loss of independence in macroeconomic policies and, consequently, the possible increase in real macroeconomic instability (due to the reduction in the number of instruments available to stabilize production, inflation or the current account). The importance of this loss will be determined by the weight of trade within the area (with respect to total trade) and by the degree of symmetry 8 of shocks (Krugman, 1991) . The growing importance of the financial system in the determination of macroeconomic policies has created the need to redefine the costs and benefits of a monetary union, with a better understanding of some of the aspects already mentioned in the traditional view and incorporating some new ones of financial character. Blending the traditional view with the financial view (and remembering the asymmetry that exists in the case of a monetary union such as dollarization) we can put together a more accurate description of the costs and benefits of replacing the domestic currency by the dollar of the United States. Benefits A first benefit is the increased credibility of the monetary policy. When a country lacks such virtue it can import it delegating its monetary sovereignty to a "credible" partner that will guarantee a certain antiinflationary discipline (Giavazzi y Giovannini, 1989; Carrarro y Giavazzi, 1990). In association with this, we have the benefit of the reduction in the risk of devaluation. With the US dollar as 9 the domestic currency, economic agents will no longer be in danger of suffering capital losses . Moreover, the reduction in the risk of devaluation implies a reduction in the cost of servicing the external debt (however, that the risk of devaluation is reduced with respect to the US dollar does not mean that the country risk disappears. The reason is that there is still the risk of default from the public sector as well as from the private one) Additionally, if dollarization reduced the devaluation risk as well as the country risk, the domestic interest rate would fall, having a positive impact on investment, growth and perceived wealth (in this last case due to the reduction in the inter-temporal discount rate). We should also stress the positive effect of dollarization on international trade between the United States and the associated country. By reducing transaction costs, dollarization induces increased trade in goods and services between the countries that do so in the same currency (Bergsten, 1999). Costs An important cost of dollarization is the reduction in the number of financial instruments available to hedge against real shocks since it implies the disappearance of nominal assets denominated in domestic currency (Helpman and Razin, 1982). In addition, when opting for dollarization the country will periodically have to buy an important amount of dollars from the Federal Reserve to compensate for additional demand for money. In this situation the 10 country cannot print domestic currency or gain interest on reserves . Furthermore, with dollarization foreign investors, which determine the country risk, know that there are no exchange instruments available to adjust relative prices and that public expenditure (in dollars) is mainly domestic payments and external debt payments (interest and capital, both in dollars). In a recession, to keep servicing its debt the government should reduce public sector expenditures in dollars or increase taxes. Thus, all of the adjustment will fall, in the short run, on fiscal measures. The perceived risk of fiscal default will increase. Also the probability that the government (having trouble to generate more revenues or reduce expenditures in dollars) will also be obliged to reduce its external debt payments will increase. In other words, dollarization would require a fiscal system even better than that of several US states since probably the associate country will not have access to the US Treasury's help through compensatory transfers, as the 11 US states now do . 8 Hit by a specific shock ε if the country is dollarized and trades little with the United States (or with other countries with fixed exchange rate with the dollar), εwill be important in relationship with intra-area trade. In such a case, there will be a need for a great change in relative prices to compensate for the shock through trade within the area. If trade is important in relationship to ε , then the change in relative prices needed to compensate for the shock will be smaller. 9 Actually, only with respect to the dollar since there might be potential losses or gains with respect to other currencies. 10 Levi Yeyati and Sturzenegger (1999) calculate that for the case of Argentina this senoriage is equivalent to 0.33% of GDP, assuming a growth rate of 5%, and inflation rate of 5% and a ratio circulating money/GDP of 3.4%. 11 Sachs and Sala-i-Martin (1991) found that in a negative shock any region of the United States gets a federal compensation of 30-40 cents for every dollar lost of income. 3 An additional specific cost would fall to the financial system which could no longer count on the Central Bank acting as a lender of last resort in a bank crisis (although potentially that role could be taken by the Federal Reserve or an ad-hoc Fund). This is not only a problem of tough and costly regulations. Even in the United States, where there are serious regulations, there have been crises where the Federal Reserve has had to intervene such as the case of the investment fund Long Term Capital (LTC) in 1998, or the crises of the regional banks in the eighties. In those countries which now have a Currency Board regime there exists an extra cost from dollarization They would have to negotiate with the US government the loss of the interest in their international reserves 12 now in deposit in the United States . Finally, another significant cost is the little importance that the United States places to its external sector in comparison with smaller countries. For that reason the dollar is allowed to fluctuate widely with respect to the Euro or Yen without the Federal Reserve or the Treasury expressing excessive preoccupation, contrary to 13 the attitude of any small country which trades mainly with the rest of the world . 4. Dollarization, nominal volatility and real volatility In the analysis of the role of the exchange rate system there are two relevant factors that we wish to highlight 1) its effects on growth, and 2) its effects on the business cycle and the volatility of the economy. With respect to the first aspect, the exchange rate regime is not a source of growth in the long run but it could reduce it if it generates excessive volatility of the economy. De Grawe (1988) shows how in a context of neoclassical growth with increasing returns to scale, it is possible that a reduction in exchange rate uncertainty, which reduces the interest rate, could increase the rate of growth. The adoption of a specific exchange rate regime could be used to: 1) reduce nominal uncertainty (or volatility), 2) control inflation 14 (Calvo and Vegh, 1993; Fanelli and González Rozada, 1998), or 3) reduce real volatility . As it has been pointed out by Helpman and Razin (1982), changing from a monetary regime with a central bank to a monetary union implies a trade-off between the benefits of reducing excessive exchange rate 15 volatility and the cost of reducing the number of financial assets available in the economy. With imperfect financial markets a flexible exchange rate regime is superior to a fixed exchange rate one because it 16 increases the efficiency with which economic agents diversify risk (Helpman and Razin, op.cit.) . A typical example of an instrument for hedging that is lost with a monetary union is the possibility of devaluation to adjust relative prices in response to shocks. When policymakers have a propensity to generate policy shocks which are expected by the population there will exist an inflationary bias whose volatility could be influenced by short run electoral objectives; in such a case a monetary rule could be optimal. Neumeyer (1998) states that a monetary union is desirable when the gains from the elimination of excess volatility generated by policy shocks ("bad" nominal volatility) exceed the costs of reducing the number of instruments available to hedge against risk. Our paper moves a step forward and complements the perspective of works already discussed by taking into account the problem of how an exchange rate regime (dollarization) affects the real volatility of the economy. We believe that the effects of dollarization on "good" and "bad" nominal volatility as well as on real volatility should be considered. Behind this idea is the problem highlighted by Poole (1970) in his pioneering paper on what the most convenient regime to reduce real volatility is depending on the source of the shocks. If shocks come from the monetary market (they affect the LM curve) then the fixed exchange rate regime seems better, while if shocks originate in the goods market (affecting the IS curve) a flexible exchange rate regime would reduce 12 Dollars that would partially return to the country as circulating money For example, in the United States an important increase in productivity has occurred in recent years, which has induced a revaluation of the dollar with respect to the Mark (Euro) and the Yen. The country that wished to change its domestic currency for the dollar would have to at least equal the productivity performance of the United States to maintain its competitive capacity. Thus, dollarization implies importing external policy shocks (such as the greater US's real exchange rate volatility) not necessarily compatible with domestic preferences. 14 Since exchange rate changes (to modify relative prices) can help to absorb shocks in the presence of price rigidities or market imperfections (Roubini, 1999). 15 From the financial point of view we can make a distinction between "good" or "bad" nominal volatility of the exchange rate whether its origin is a "political" shock or a real shock (in preferences, in resources or in productivity). This last type of volatility is functional to the reallocation of factors and resources in an efficient way (Neumeyer, 1998). 16 An important assumption in this case is that there are adequate instruments for risk diversification within the countries but not amongst them. This is consistent with a context where there is a preference for domestic assets, or where there are no enforcement authorities at a supra-national level with fiscal institutions that could help diversify risk as national governments do. 13 4 the volatility of output. From this section we can conclude that the effect of the exchange rate regime on real fluctuations is relevant for two motives. First, the greater the real volatility, the greater the domestic price changes the economy will require and, thus, the greater the advantages of a flexible exchange rate regime that admits certain nominal volatility ("good" volatility in the sense expressed by Helpman and Razin, 1982). In the second place, the greater the real volatility, the bigger the country risk implied in domestic assets. In the next section we discuss this last proposition in depth. 5. Volatility and country risk When there are no restrictions on the mobility of capital and agents are neutral towards risk, the condition of uncovered interest rate parity has been widely used to evaluate the possibilities of arbitrage between different financial markets. Under these conditions any deviation is a white noise, unpredictable and of transitory character. Kaldor (1939) stated that in the activity of arbitrage it is necessary to allow for a risk premium that takes into account the problem of uncertain expectations and that more dispersion in expectations should imply a greater risk premium. In recent decades the importance of country risk to justify observed interest rate differentials between similar assets from different countries has been pointed out 17 (Dooley, 1995). We can thus obtain the risk-adjusted interest parity condition : δt = rt − rt* − E∆ee − ut (1) where δ is the difference in the interest rate between two assets of the same maturity and risk characteristics, r is the domestic interest rate, r* is the international interest rate, E∆ ee is the expected rate of devaluation and u is an IID random variable. Roubini (1999) states that the country risk (δ) can be interpreted as the risk of default of domestic assets. Since the 1994 Mexican crisis and the successive Asian, Russian and Brazilian crises, in the case of the so called emerging countries, attention has been focussed on the role of the country risk in explaining the abrupt dismissal of exchange rate regimes and/or the recession that came with the process of absorption of negative shocks. Some authors such as Calvo, Leiderman and Reihart (1993) and Calvo (1998) study the incidence of contagion effects on country risk. Others such as Avila (1998) and Rodríguez (1999) focus on the negative effect of country risk on the rate of change of output. However, what does not seem to have been analyzed with equal depth is the inverse relation. That is, what is the effect on the country risk of increased expected output volatility? For example, output volatility could be caused by growing efforts to reduce the nominal volatility with a rigid exchange rate regime (such as dollarization). The relationship between volatility and country risk is highly intuitive: greater expected real volatility of output implies greater uncertainty as regards the profitability of investment and economic agents' consumption plans. This raises doubts as to the possibility of recovering invested capital or with respect to expected profits from holding domestic assets. Greater uncertainty will make agents require bigger returns from domestic assets in comparison with similar assets in countries with less real volatility. To determine which theoretical position is the best answer to this problem, we check in the case of Argentina the relationship between the volatility of the business cycle and the country risk. Based on different econometric methodologies, we found that there exists a positive (and very significative) relationship 18 between these variables, with increased volatility of the business cycle increasing the country risk . Based on these results we may state that the effect of dollarization on country risk (through its impact on real volatility) will depend on: 1) The exchange rate regime of the leader (in this case the United States) vis à vis the rest of the world. 2) The characteristics of the business cycle of the leader, since he is a source of shocks from which it is not 19 possible to insulate . 17 Edwards (1999) applies a similar formula, that includes the equivalent rate when there are capital controls, for the discussion of equilibrium interest rates differentials. 18 In the section 2 of the appendix we present the results for cross-correlation coefficients and linear regression model between these variables. 19 The associated country could insulate from this shocks only with its fiscal policy. However, a country can use its fiscal policy if and only if this policy is not pro-cyclical by design, something that would invalidate it for this goal. See on this Carrera, Pérez and Saller (1999), and IDB (1995). 5 In 1) the exchange rate regime of the United States could insulate the associated country (AC) from shocks from the rest of the world or generate additional shocks (i.e., the appreciation of the dollar caused by a shock of productivity in the USA would generate a change in relative prices in the AC). In 2) it is stressed that the business cycle of the United States is a source of external shocks and depending on how fluctuations are transmitted they could amplify or reduce the cyclical volatility of the country associating with the dollar. To move forward in our analysis we concentrate on how fluctuations in the Unites States' output are transmitted and what effect they have on the country that associates as a follower in this asymmetric monetary union (dollarization). 6. How does the United States’ cycle transmit in the context of an asymmetric monetary union? The business cycles of the different countries, understood as the variations of output around its trend, do not relate directly but through channels that transmit shocks from one economy to the other. In the case of an asymmetric relationship big country-small country the transmission of the effects of the business cycle originated in the main economy to the different small countries occurs mainly through the transactions of goods (and services) and of financial assets (Canova and Ubide 1997; Schmitt-Grohé, 20 1998) . With the aim of simplifying the theoretical and empirical analysis, we may decompose the channels of transmission into two great groups: the financial channel and the trade channel. The financial channel is related to the effects of the international interest rate on the level of capital flows to the emerging economies. This effect could be very significant as regards the size of fluctuations in the periphery for two reasons: 1) its determination is dominated by the economic conditions of the main center and thus they do not necessarily respond to the counter-cyclical needs of the emerging countries; 2) the high level of dependence on external savings by the emerging economies makes them very vulnerable to the perturbations in the international interest rate (Calvo, Leiderman and Reinhart, 1993). In the trade channel, on the other hand, the effect of fluctuations in the business cycle of the leading economy (the United States) is transmitted through the movements in the trade flows (due to changes in quantities as well as in the terms of trade). The relative size of each channel will indicate the magnitude of the effect on the economy hit by the shocks. If the channel has a very small magnitude in relationship to the economy under study, shocks coming through it will have only moderate effects on the cycle. To summarize the previous analysis we present a simplified representation of the channels of transmission 21 of cyclical fluctuations from USA to the AC (Figure 1) . 20 For a survey on the correlation of the macroeconomic aggregates in the countries of the OECD see Blackburn and Ravn (1991) and Backus et at. (1992) for a presentation in the context of the Real Business Cycle theory. 21 The signs and relative importance of the different channels will be analyzed in the empirical section of the paper using vector error correction models with cointegrating relationships, and different estimations of the cyclical correlation between the variables of interest. 6 Figure 1 Channels of transmission of the United States' business cycle Domestic Shocks Bilateral Trade USA - AC Trade Balace AC-USA Extra union trade shocks GDP USA i* GDP AC i Domestic Shocks Extra union finatial shocks where AC is the country associated to the dollar, i is the interest rate that prevails in AC and i* is the interest rate determined by the Federal Reserve. Each country suffers from domestic shocks but the small associated country also receives the influence of external shocks that are transmitted from the main center (the United States). The two channels of transmission are the financial channel (represented by the level of the United States interest rate) and the 22 trade channel (represented by trade between the two countries) . This effect can be decomposed into two stages represented by the complete lines: a) the impact of United States' imports and of the Federal Reserve's interest rate on trade from AC and on the interest rate of AC, respectively, and 2) the effect of this later variables on the GDP of the associated country. From the perspective of AC, the economic intuition behind this simplified representation of analysis for the transmission of economic shocks is as follows: the United States' economy transmits its shocks through the trade channel and through the financial channel. When the USA's GDP is hit by a positive shock, two simultaneous processes begin: a) United States imports increase (affecting positively the GDP of AC through the trade channel), and b) the Federal Reserve increases the interest rate to slow down the economy to avoid over-heating (transmitting a shock through the financial channel that will affect negatively AC). These hypotheses which refer to the mechanism of transmission of the business cycle may be formalized in the following expressions: ∂M USA− AC >0 ∂GDPUSA 22 (2) The dotted lines synthesize the effects of the different channels on the GDP of the country associating to the dollar. 7 ∂i * >0 ∂GDPUSA (3) ∂X AC −USA =1 ∂M USA− AC (4) ∂i >0 ∂i * (5) ∂GDPAC >0 ∂X AC −USA (6) ∂GDPAC <0 ∂i (7) where X AC −USA represents exports from AC to the United States (symmetrically, M USA− AC represents United States' imports coming from the country dollarizing its monetary system). In section 2 of the appendix we present the econometric estimations and/or bibliographic references that provide empirical support for the hypothesis of equations (2), (3) and (5). Equation (4) derives from an 23 24 accounting identity , while the hypotheses contained in equations (6) and (7) will be confronted in the empirical section of the paper when we apply this analytical framework to the case of Argentina. 7. The problem of the policymaker confronted with the alternative of dollarization The adoption of a more rigid exchange rate system such as dollarization could reduce real volatility if it acted as an automatic stabilization mechanism of the economy. This is a very important issue since a risk averse policymaker will prefer a more stable growth rate since this reduces the country risk perceived by (also risk averse) investors. To make an evaluation of the aggregate effect of dollarization, we use a unified framework that takes into account the different effects (from the financial point of view) that are associated with this asymmetric monetary union. We want to specify which are the necessary conditions to ensure that dollarization will increase social welfare. We assume that the policymaker wants to minimize a social loss function that represents the external and financial fragility of the country (Fanelli and Gonzalez Rozada, 1998), where the control variable is the degree of rigidity of the exchange rate system (d). L d = f [ RD ( d ), H ( d ), RP (σ GDP . AC ( d )) ] (Policymaker's social loss function ) (8) where RD(d ) is the risk of devaluation, H (d ) represents the number of financial instruments available to compensate for (or cover against) real shocks to the economy (Helpman and Razin, 1982), RP (σ GDP. AC ( d )) is the country risk (a positive function of the real volatility of domestic GDP) and d is a continuous variable 25 representing the degree of rigidity of the exchange rate system . Differentiating the loss function with respect to d (under the assumption that, for example, this change takes the form of the conformation of an asymmetric monetary union such as dollarization) we obtain equation (9). ∂RP (σGDP. AC ( d )) ∂L ∂RD ∂H = f RD + fH + f RP ∂d ∂d ∂d ∂d (9) We find that the result will depend, as we expected, on the assumptions (to be tested econometrically) made about the signs of the different coefficients. With respect to the signs of the coefficients involved we assume that: 23 24 It is obvious that X AC −USA = M USA− AC . We will confront the hypotheses of equations (16) and (17) (to be presented in the following pages) which put together those in equations (4) and (6), and (5) and (7), respectively. 25 The sequence would be flexible exchange rate, "administered" or "crawling" exchange rate, flotation bands, traditional exchange rate fix, currency board and dollarization or, more generally, the use by a country of another country's currency. 8 f RD > 0 , (10) fH < 0 , (11) f RP > 0 . (12) The sign of (10) is positive in as much as the social loss increases with the increase in the risk of devaluation. This risk characterizes the "bad" volatility that is related to the inflationary bias of the system when there is a discretionary monetary policy. With respect to (11) its sign depends on the results from Helpman and Razin (1982) where a greater number of nominal financial instruments reduces the social loss since it allows for better risk diversification by allowing the fluctuations in the exchange rate that act as an instrument for the diversification of real risk. Finally, equation (12) implies that, as with the risk of devaluation, the social loss increases when the country risk increases. With respect to the rest of the partial derivatives from the different sources of social loss involved with respect to dollarization, we assume the following signs: ∂RD <0 ∂d (13) ∂H <0 ∂d (14) ∂RP (σ GDP. AC ( d )) ∂RP ∂σ GDP. AC > = . =0 ∂d ∂σ GDP. AC ∂d < (+ / − ) (15) ( +) In (13) we state that a movement towards a more rigid exchange rate regime such as dollarization reduces 26 the space for independent policies by the central bank thus eliminating the risk of devaluation . In equation (14) we simply indicate that dollarization reduces the set of available nominal financial instruments in the economy. The central point in our analytical framework relates to the sign of equation (15) that will determine the final result of equation (9). The central problem is to determine the effect of dollarization on the country risk. To find this result we need to remember our previous discussion on how the United States' business cycle is transmitted through the financial channel (FC) and the trade channel (TC). Combining equations (4) and (6) we obtained the expected effect of a shock transmitted through the trade channel on the associated country's GDP: ∂GDPAC >0 ∂M USA−AC (16) In a similar fashion, combining (5) and (7) we may find the expression that summarizes the effect of the 27 financial channel on AC's GDP : ∂GDPAC <0 ∂i * (17) Without loss of generality we present a functional form in which real volatility σGDP. AC depends on the relationship between the business cycles of the United States and the associated country, and the relative importance of each channel. Thus we have the following expression: 26 Levi Yeyati and Sturzenegger (1999) show that even dollarization could be reversed. They state certain conditions under which there are perverse incentives for the policymaker in a country that is receiving the dollars it needs to substitute its domestic money supply (sharing the benefits of senioriage). They show that under certain conditions, the policymaker may renounce the compromise and reinstate the domestic currency. However, this seems more a case of theoretical importance than of practical significance given the high punitive power of an agent such as the Federal Reserve. 27 In the empirical section we test the validity of the hypothesis of equations (16) and (17). 9  ∂σ GDP. AC IMPTC IMPFC = g  VDSTC + VDSFC ∂d IMP IMPFC  TC with g ′ > 0   .ρ GDP  AC / USA    (18) g´´< 0 where g is a monotonically increasing (at decreasing rates) function in the argument, VDS TC and VDS FC indicate, respectively, the participation of the trade channel and the financial channel in the volatility of AC's product, IMPi represents the effect of each channel in the product of AC (thus, the ratio IMPi/|IMPi| with i= (FC, TC) allows us to obtain the sign of the effect of a change in the variable that represents each channel of 28 transmission ) and ρGDP indicates the correlation between the cycle of AC's GDP and the cycle of United 29 AC / USA States' GDP . Assuming the hypothesis of equations (2), (3), (16) and (17), equation (18) implies that: 1) In the case of the trade channel (TC), when the two economies are in an expansionary phase of the cycle (outputs are positively correlated), an increase in the demand in the United States produces an increase in exports from AC towards the USA. Since exports are a component of aggregate demand this works as an additional external positive impulse that gives an additional pull to the business cycle in AC. In this case the trade channel increases real volatility. On the contrary, if the cycles are negatively correlated, when AC is in a downward phase of the cycle the United States is expanding. In this case an increase in external demand for goods from the United States implies an increase in exports from AC and, thus, a positive impulse in its GDP. In this case the trade channel reduces real volatility. 2) With respect to the financial channel, let us assume that the two economies were positively correlated and both expanding. In this case the increase in the interest rate by the Federal Reserve to avoid over30 heating the USA's economy would produce a similar effect in the economy of AC . In this way, the cycle would be contained, reducing the range of fluctuation of the growth rates in AC. On the contrary, if the two economies were negatively correlated, that is when the United States is growing the associated country is in recession, the increase in the Federal Reserve's rate would increase the downturn in the AC's economy. Table 1 resumes the previous discussion. With synchronized cycles ( ρGDP >0) the financial channel (FC) AC/ USA reduces the volatility of the cycle and the trade channel (TC) increases it. On the contrary, with cycles negatively correlated, the FC increases real volatility and the TC reduces it. Table 1 Change in the volatility of AC GDP Channel of transmission TC FC ∂GDPAC >0 ∂MUSA−AC ∂GDPAC <0 ∂i * ρGDP Cyclical correlation AC / U S A >0 ρGDP AC/ U S A <0 A B ↑ real volatility ↓ real volatility C D ↓ real volatility ↑ real volatility From this framework of analysis we may establish the following two propositions: Proposition 1: Assuming the usual mechanisms for the transmission of the business cycle in a center31 periphery framework , dollarization will reduce real volatility and thus the country risk if and only if the following conditions are fulfilled: a) if correlation between business cycles is 28 By dividing the value of the coefficient by its absolute value we exclusively keep the sign which is adequate since our objective is to isolate the result from the size of the effect. 29 VDS T C and VDS FC will be proxied in the empirical section by the proportion of the variance of the output of AC that is explained by the trade and financial channel, respectively. The values for IMPT C e IMPFC will be obtained from the responses of AC's output to a shock in the trade and financial channel, respectively. Finally, ρ will be proxied through different estimations for the coefficient of cyclical GDPAC / U S A correlation between the United States' and AC's output. 30 For an analysis of the negative association between the United States' interest rate and the level of activity in Latin American countries see Calvo, Leiderman and Reinhart (1993), Frenkel (1998), and Roubini (1999). 31 Hypotheses contained in equations (2), (3), (16) and (17) which will be checked and ratified in the empirical section and in section 3 of the appendix . 10 positive, the financial channel should dominate the trade channel. b) if the correlation between the cycles is negative, the trade channel should dominate the financial one. In terms of equation (18) we have the following alternatives when we combine the possible results for each 32 of the free variables of the equation : VDS TC > VDS FC and ρ AC / USA > 0 (19) VDS TC < VDS FC and ρAC / USA > 0 (20) VDS TC > VDS FC and ρ AC / USA < 0 (21) VDS TC < VDS FC and ρAC / USA < 0 (22) VDS TC = VDS FC (23) ρAC / USA = 0 (24) In the case of expressions (19) and (22), dollarization will increase the real volatility of the GDP of AC ( σ GDP. AC ). If, on the contrary, we verify that the cyclical behavior and the relative importance of the different channels correspond to expressions (20) or (21), dollarization will allow the associated country to import a monetary policy that will act as an automatic stabilizer of its economy, reducing σ GDP. AC . Finally, if the channels have the same relative importance (equation 23) or if cyclical correlation is not significantly different from 0 (equation 24), then dollarization induces no effect on the associated country's real volatility. The economic intuition in proposition 1 can be presented clearly through the following two examples: Case A (equation 19) If both countries were in recession when their cycles are synchronized ( ρ AC / USA > 0 ) and the trade channel dominates the financial channel (VDS TC > VDS FC ), the fall in exports of AC to the United States (due to reduced USA's demand) would accentuate AC's domestic recession. Since the financial channel is of little importance the counter-cyclical policy that the Federal Reserve could be practicing in the United States would not be enough to compensate the volatility amplifying effect of the trade channel. In this case, dollarization implies resigning an instrument (such as the exchange rate policy) that could act as a stabilizer of the business cycle reducing real volatility. Case B (equation 22) If the United States' economy is expanding while AC's economy is in recession, the increase in the demand for AC's exports could smooth AC's recession. However, since the financial channel dominates, the countercyclical policy of the Federal Reserve increases the downturn in AC (since AC's interest rates will have to increase there too). Once again, dollarization implies losing the possibility of using a domestic countercyclical policy either to practice expansive policies or to compensate for the increase in the Federal Reserve's rate. The economic intuition behind the other alternatives (equations 20, 21, 23 and 24) can be easily derived from the previous examples. Proposition 2: Dollarization will improve social welfare if the weight given by the policymaker (society) to the reduction in the aggregate risk (devaluation risk plus country risk) is greater than the loss of social welfare due to the reduction in the number of available instruments to cover for risk. 32 It is important to remember that from the assumptions in equations (16) and (17), we already know the results for IMPT C (positive) and IMPFC (negative), so that the only free variables are VDS T C, VDS FC and ρ . GDPAC / U S A 11 ∂L < 0 si ∂d   ∂RD +  ∂RP ∂σ  ∂H f RD f RP  ∂σ  < − fH ∂D  ∂d 144 4442444 ∂d 44 3 GDP. AC (+ ) ( +) ( −) 14 424 4 3 ( +) GDP. AC ( +) (+ / − ) ( + / −) (25) ( −) ( −) 14243 (− ) The intuition is that given a certain level of loss due to the disappearance of an instrument for hedging the greater the reduction in the country risk, the smaller the reduction needed in the devaluation risk for dollarization to be welfare improving. This framework for the analysis of the net benefits of dollarization is sufficiently general to be applied to the different countries. In the second part of this paper we find the signs and dimensions corresponding to the case of Argentina, focusing the analysis on the relationship dollarization - real volatility - country risk. 8. Dollarization, real volatility and country risk. An application to the case of Argentina In this section we develop an empirical application of the analytical framework presented in the previous sections to assess the potential impact of dollarization on the real volatility of Argentina’s economy. The selection of Argentina as a case of study relates to the fact that, in this country, dollarization has been 33 subject to intense debate in political as well as academic spheres lately . Moreover, the president of Argentina's Central Bank has presented an official proposal to implement a Treaty of Monetary Association with the United States (Pou, 1999). The interest of certain Argentine economists in implementing dollarization relates to the lack of credibility of Argentina's monetary policy, as a result of years of excessive exchange rate volatility and inflation rates that reached the 200% monthly. In 1989 the first (constitutional) presidential succession in four decades occurred, in the midst of an unprecedented economic crisis, the most significant manifestation of the crisis was hyperinflation. After a number of stabilization plans failed in the early nineties, the government relied on a radical solution to the fiscal problem and fixed the exchange rate as a means to stop inflation. But the government needed an instrument that would help it gain credibility. Besides the fixation of the exchange rate, Argentina established, by law, a compromise not to devaluate the currency and to fully back the money base with its hard currency reserves (Convertibility law, a currency board regime). As a mechanism for coordinating expectations and reducing inertial inflation, indexing of wages and prices was prohibited. In addition the reduction of trade 34 restrictions was used as a price control mechanism for tradable goods . The implementation of these policies was complemented with others of structural character such as privatization of public services and State reform (reduction in the number of public sector workers, decentralization of basic services, etc.). After 9 years of virtually null inflation and strong growth, the marginal benefits of these policies seem to be fading out. In relationship with the exchange rate regime, there are disputes between those who propose the devaluation of the exchange rate to compensate for a number of recent negative real shocks and those who believe that the right direction is exactly the opposite, that is to go deeper into the Convertibility, dollarizing the economy. For these motives, the analysis of the impact of dollarization on the real volatility of the Argentinean economy constitutes an empirical application of great relevance to the decisions of economic policy that relate to the monetary system of this country. The empirical analysis consists of two stages. First, we present a vector error correction model (VECM) to examine the effect and relative importance of the trade and financial channels in Argentina. Second, we examine the correlation between the business cycle of Argentina and the cycle of the United States. The objective of this second part is to obtain an appropriate estimation of the coefficients of cyclical correlation to determine (together with the effects and relative importance of the different channels) the impact of dollarization on social welfare through its effect on country risk (which is a positive function of real volatility). 33 Other Latin American countries have been discussing the subject lately. For example, several working papers from researchers at the Central Bank of Costa Rica take into account the idea of dollarization in Costa Rica seriously (see for example Ramos et at, 1999). Recently, Ecuador's government has announced the intention of abandoning the Sucre, their local currency, in favor of the US dollar. The idea resulted in tremendous upheaval (which included a failed coup-de-etat) within the country. 34 Few countries in the world have a regime such as this, amongst them (besides Argentina) Hong Kong, Estonia, Lithuania and Brunei. For a comparative study of the current experiences see the paper by the IMF on Currency Board in Baliño et at (1999). 12 8.1. Effect and relative importance of the different channels of transmission of the business cycle. A VEC model approach 35 Following the traditional methodology to analyze the structure of the different shocks that hit the economy , we build the vector error correcting model (VECM) to describe the way in which shocks are transmitted from 36 the United States to Argentina . VAR models are used in the prediction of the series included in them and for the identification of the different kinds of shock, affecting the economies. Our work makes use of the two tools derived from VAR models: impulse-response functions and variance decomposition procedure. To use the impulse-response functions and the variance decomposition procedure it is necessary to identify the shocks for each and every variable in the system. In more general terms, n(n-1)/2 restrictions are needed to exactly identify the model (where n is the number of variables in the model). For that purpose, one methodology that provides these restrictions is the Cholesky decomposition which imposes that the matrix A(0) (which incorporates the contemporaneous effects of the variables) be triangular 37 inferior . Different authors have criticized the arbitrary methodology of imposing restrictions of identification on the Cholesky decomposition, indicating, for example, that the results in most cases (when there is correlation 38 amongst the residuals of the equations) are very sensitive to the order in which the variables are included . Alternative solutions have appeared. Using the general structure of the VAR models, changes are introduced in the identification restrictions. Among them, the developments by Blanchard and Quah (1989) and Johansen (1991, 1995) stand out using long run restrictions to identify the different models. However, there are noticeable differences as regards the reasons why restrictions are introduced in each methodology. While Blanchard and Quah (based on the supposition of a vertical aggregate supply curve in the long run) determine that demand shocks will not last, Johansen’s methodology takes the long run restrictions from the data generating process without imposing ad-hoc behavioral restrictions on the different markets. 39 In this paper we use Cholesky decomposition to find short run identification restrictions and Johansen’s methodology to estimate long run relationships without having to impose a priori restrictions. The structure of the model can be easily explained through the following example of a VEC with n variables 40 and one lag for each variable . Let: z t = Γ1 z t −1 + εt where (26) z t = the (nx1) vector [ z1t , z 2t , z 3t , z 4 t ,........., z nt ] of variables in the model εt = the (nx1) vector [ε1t , ε2 t , ε3t , ε4 t ,........., εnt ] of gaussian errors Γ1 =an (nxn) matrix of parameters. Subtracting z t −1 from each side of (26) and letting I be an (nxn) identity matrix, we get, ∆z t = −( I − Γ1 ) z t −1 + εt , or ∆z t = πz t −1 + εt (27) 35 See Sims (1980), Blanchard and Quah (1989), Johansen and Juselius (1992), and Amisano and Giannini (1997) for theoretical and empirical applications in which VAR or VEC models are used to identify the different shocks hitting an economy. 36 In recent years there has been a great number of papers which study the transmission of the international business cycle using VAR modeling for the empirical analysis. Amongst the most interesting papers in this field of research we recommend: Calvo, Leirdeman and Reinhart (1993), Chamie, DeSerres and Lalonde (1994), Canova (1995a), Horvath, Kandil and Sharma (1996), and Schimitt-Grohé (1998). 37 For further detail see Hamilton (1994). 38 See Enders (1995). 39 Knowing that under this kind of decomposition the identification of shocks is very sensitivity to the order in which the variables are included in the model, we also develop a sensitivity analysis for the results that includes 6 different orderings for the variables. 40 Following the description presented by Enders (1995). 13 where π is the (nxn) matrix − ( I − Γ1 ) t and πij denotes the element in row i and column j of π . πij is equal to 0, the rank of the matrix π is 0 and (27) is equivalent to an n-variable unrestricted If each VAR in first differences. On the other extreme if equations: π is of full rank the long run solution to the system is given by the n independent π11 z1t + π12 z 2t + π13 z 3t + L + π1n z nt = 0 π21 z1t + π 22 z 2 t + π 23 z 3t + L + π 2 n z nt = 0 . . . . . . . . . . . . . . . . . . πn1 z1t + π n 2 z 2 t + π n3 z 3 t + L + πnn z nt = 0 In this case none of the series has a unit root, and the VAR may be specified in terms of the levels of all of the series. If there are r<n vectors of cointegration, the VAR should be re-expressed in first differences with the inclusion of the r independent error correction mechanisms that establish the long run relationships between the variables. Assuming that r=1, each sequence ∆z1t as: {z } can be written in error correction form. For example, we may write it ∆z1t = π11 z1t −1 + π12 z 2 t −1 + π13 z 3t − 1 + L + π1n z nt −1 + ε1t or, normalizing with respect to z1t −1 : ∆z1t = α1 ( z1t −1 + β12 z 2t −1 + β13 z 3t −1 + L + β1n z nt −1 ) + ε1t where (28) α1 determines the speed of adjustment to a long run dis-equilibrium, while the β1i give us the coefficients which determine the long run relationship. These results remain unchanged if we formulate a more general model by introducing the lagged first differences of each variable into each equation. In such fashion we obtain the following expression that includes the n equations of the model (assuming that there exists only one vector of cointegration, that is r=1): n k ∆z it = αi ( z1t −1 + β12 z 2 t −1 + β13 z 3t −1 + L + β1n z nt −1 ) + ∑∑ψij ∆z it − j +ε1t (29) i =1 j =1 where ψij is a (nx1) vector of parameters for equation i and lag j. Equation (29) represents a VEC model with n variables, one cointegrating vector and k lags for the variables in first differences. We will use this type of VECM to evaluate the effect and relative importance of the financial and trade channels in the transmission of the business cycle from the United States to Argentina. Next we present the main characteristics of the model and later on we show the most important results. The model 41 We develop a vector error correction model with three variables (Federal Reserve Interest Rate 42 (FEDRATE), imports of the United States from Argentina (IMPOUA) and the Industrial Production Index of Argentina (IPIARG)), and an intercept in each equation. 41 42 The sample period for the model is 1991:4 – 1999:10 , with monthly data. Equivalently, exports of Argentina to the United States. 14 43 44 The variables are in logs, seasonally adjusted and in first differences . FEDRATE and IMPOUA represent 45 the financial and trade channels respectively . The IPIARG is taken as an approximation of Argentine GDP. As we can see in the section 1 of the appendix, we use the Industrial Production Index instead of the actual GDP because there are no reliable estimations for Argentina’s GDP on a monthly basis and because the correlation coefficients between these variables is extremely high. The model is thus defined as follows: k ∆. zt = ∑ Ψi ∆. zt − i + Π. zt −1 + µt + εt (30) i =1 where z t −1 = the (3x1) vector [FEDRATE , IMPOUA t− t −1 , IPIARGt −1 ] , ∆z t = the (3x1) vector [∆FEDRATE , ∆IMPOUA , ∆IPIARG ] , t t t εt = the (3x1) vector [ε1t , ε 2t , ε3t ] of uncorrelated, homocedastic, gaussian errors, µt = the (3x1) vector of the deterministic components, Ψi =an (3x3) matrix of parameters, and Π is the (3x3) matrix of rank r (to be tested) which contains the parameters of the cointegrating vectors. The next step consists in verifying the conformity of the model with the data generating process evaluating the order of integration of each variable, the rank of Π and the optimal lag length. Unit root test For each variable (in levels and in first differences) in the model we perform the ADF (Dickey and Fuller, 46 47 48 1979) and Phillips-Perron (1988) tests to detect the presence of unit roots in the series . In table 7 in section 4 of the appendix we present the results of the different tests for a confidence level of 95%. We verify that almost all variables are I (1) (integrated of first order). There is a certain contradiction for the 49 variable IPIARG . However, the results that indicate that the series is I (1) seem more robust since 83% of the tests for this variable (5 out of 6 different specifications for the ADF test and the Phillips-Perron test) state that its data generating process would be correctly represented by a random walk. Since every variable can be considered I (1) we fulfill the first necessary condition for the construction of a 50 VECM . The second necessary requisite to build the model is that the rank of the matrix of the cointegrating vectors should be greater that 0 (zero) and less than n. For our model the rank of the matrix should be equal to 1 or 2. Tests for the optimal lag length. According to Canova (1995a) "the trade-off between over-parametrization and oversimplification is at the heart of the selection criteria designed to choose the lag length ". There are different selection criteria to determine the optimal number of lags in VEC models. In this paper we use some of the more traditional such a Akaike criterion (Akaike, 1973), Schwarz criterion (Schwarz, 1978) 43 Through X-11 ARIMA method. For further details see section 1 of the appendix. It is convenient to remember that in the error correction mechanism(s) the variables appear in levels and lagged one period. 45 In section 1 of the appendix we present the reasons for the sample period and for the variables included in the model. Also, we show the characteristics of the series (periodicity, transformations, etc.), the econometric software used (and the procedures) and the source of the information for each variable. 46 The number of optimal lags for each specification was obtained following Akaike criterion (Akaike,1973). 47 For this test we take the truncation lag recommended by Newey - West (1994) for monthly series. 48 We checked the following three specification for the deterministic component in both tests: 1) unrestricted, which includes time trend and intercept, 2) idem 1, but without the time trend, and 3) restricted, without any deterministic component. 49 Even when the variable is I (1) for almost all specifications (5 out of 6) for the ADF test as well as for Phillips-Perron test, when we include the time trend and the intercept in the equation, the Phiillips-Perron test indicates that the series is I (0). 50 Recently, Pessaran et al. (1999) have proposed a new approximation to test the existence of long run relationships that is independent of the order of integration of the regressors. 44 15 51 and the Modified Likelihood Ratio Test (MLR test) (Sims, 1980) . In all cases, the criteria select the number of lags that minimize a loss function that has implicit the trade-off described by Canova (1995a) in the previous reference. The main distinction between the different criteria is the relative weight given to the explicative power in relation to the degrees of freedom. 52 In tables 8 and 9 (included in section 5 of the appendix), we present the results of these tests . The optimal number of lags for our model is 2, 4 and 7 for Schwarz criterion, Akaike criterion and the MLR test, respectively. To overcome this contradiction we perform an exhaustive analysis of the residuals through the joint evaluation of tables 10, 11 and 12 (also presented in section 5 of the appendix), which show the results for autocorrelation, heterocedasticity and normality tests on the errors. Clearly, the best specification includes 7 lags in each variable, since it presents residuals that are less autocorrelated, with a distribution function that is closer to a normal distribution. Establishing the number of cointegrating vectors The third step in the construction of the VECM is to identify the number of cointegrating vectors (the rank of the Π matrix) to establish the long run relationships between the variables. Postulating linear trends for the series, the results of the Johansen trace test (1988) (when we assume the existence of an intercept in the error correction mechanism) show the existence of 1 (one) cointegrating 53 vector . Table 2. Cointegration Analysis Trace Test for cointegrating vectors H0 (number of cointegrating vectors): Eigenvalue Likelihood Ratio 5% Critical value 1% Critical Value None * 0.173 33.47 29.68 35.65 At most 1 0.120 13.79 15.41 20.04 At most 2 0.005 0.57 3.76 6.65 Note: *denotes rejection of the hypothesis at 5% significance level Having determined the specification for the model that best adapts to the joint data generation process, we proceed to evaluate the long run relationship and to obtain the impulse-response functions and the variance decomposition. Analysis of the long run equation From the estimation of the autorregresive vectors model in an error correction representation, with one cointegrating vector and 7 lags, we obtain the following long run relationship between the variables (normalized for IPIARG): IPIARGt = 2.40 + 0.70*IMPOUA t - 0.77*FEDRATE t + ε t (t-values) (3.39) (2.94) (-2.20) (31) As we can see, every coefficient is significative at a 95% confidence level and have the expected signs, according to equations (16) and (17) in our analytical framework. However, to test the robustness of this long run relationship, it is necessary to evaluate two additional 54 elements. First, the residuals of the error correction mechanism, presented in Figure 2, are stationary . 51 For further description of these tests, see Hamilton (1994). Obtained, as usual, from an unrestricted VAR which includes the three variables of the model (FEDRATE, IMPOUA and IPIARG) expressed in levels. 53 In section 7 of the appendix we present the results of the sensitivity analysis in which we evaluate the robustness of the estimations when altering certain assumptions of the model, amongst them the assumption of a linear trend in the series and the intercept in the error correction mechanism. 54 Following Charemza and Deadman (1997), the existence of a cointegrating relationship between the variables requires as a necessary but not sufficient condition that the variables have the same order of integration (d). Besides, it is necessary that there exists a linear combination of these variables integrated of order d-1.Since in our case all five variables are integrated of order 1 (d=1), fulfilling the first condition, for the existence of a long run relationship we should find that a linear combination of the variables is integrated of order 0 (d-1=0). 52 16 Second, the coefficients that represent the speed of adjustment for each equation with respect to the long run error are negative and less than 1 (one) in absolute value. This implies that the system will not be explosive. Figure 2. Residuals of the Error Correction Mechanism included in the VECM 0.6 0.4 0.2 0.0 -0.2 -0.4 92 93 94 95 96 97 98 99 Residual from EC mechanism In table 13 (included in section 6 of the appendix) we present the results of the ADF and Phillips-Perron tests (with three different specifications for the deterministic component used in the equation of each test) according to which the residuals of the error correction mechanism are I (0). Analyzing the different coefficients of the VECM we find that the coefficients that pre-multiply in each equation of the error correcting mechanism (which may be interpreted as the αi in equation 29) fulfill the second requisite established above. More precisely, the coefficients of adjustment with respect to the deviation of the long run relationship for the equations of FEDRATE and IPIARG are -0.072 and -0.056, respectively. For the equation of IMPOUA the coefficient of adjustment is not significantly different from 0 at a 95% confidence level. These results show the robustness of the long run relationship between IPIARG and the different channels of transmission of the business cycle and, at the same time, they present preliminary evidence in favor of the hypothesis established in equations (16) and (17) which will be tested in the next section, with the analysis of the impulse-response functions. Analysis of the effects of the trade and financial channel on IPIARG through impulse-response functions Examining the response of IPIARG to a shock of one standard deviation on FEDRATE and on IMPOUA we find the signs needed to determine the impact of the financial channel and the trade channel on Argentina’s economy. In figure 3 we see that the different channels affect IPIARG according to the hypothesis established in equations (16) and (17). The financial channel has a negative effect on IPIARG, while the trade channel has a positive effect on it. 17 Figure 3. IPIARG response to one standard innovation in: 100.5 100.0 99.5 99.0 98.5 0 2 4 6 8 10 Financial Channel 12 14 16 18 20 22 24 Trade Channel Note 1: For better illustration we present the results normalized at 100 in i=0. Note 2: The standard deviation in the trade channel is of 13% (approximately U$S300 million) while in the financial channel the shock is of 2.5% (approximately 13 basic points). It is important to highlight, the fact that there are important differences as regards the intensity and duration of these effects. As it can be seen in the previous figure, while the effect of shocks transmitted through the trade channel tend to disappear in the long run, shocks transmitted through the financial channel have a permanent effect. This difference can be expressed more clearly by means of the following calculation: Knowing that the monthly value added by Argentina’s industry is slightly more than U$S 3,200 million and 55 using a monthly effective rate of 0.5% , we calculate (for a 24 month period) the accumulated loss resulting from a shock of one standard deviation in FEDRATE and the accumulated gain of a shock of one standard deviation in IMPOUA. In both cases, we calculate for each period the difference between the value of IPIARG after the shock and the value that IPIARG would have had if the shock had not occurred. Then we capitalize these differences to obtain that accumulated loss and gain respectively. A shock in FEDRATE results in an accumulated loss of U$S 1,200 million while a shock in IMPOUA produces a U$S 440 million 56 dollar gain . Next, we analyze the variance decomposition of a shock on IPIARG to obtain further results as regards the relative importance of the different channels of transmission of the business cycle from the United States to Argentina. Relative size of the different channels. A variance decomposition approach The variance decomposition will allow us to estimate the relative importance of the trade and financial channels. Figure 4 shows that the results of the variance decomposition are similar to those coming from the analysis of the impulse-response functions. The financial channel dominates the trade channel and the differences in explicative power between both channels increases in the long run. 55 0.5% is the approximate monthly effective rate taking into account the Federal Reserve rate was on average during the period 6%. Remember that the accumulated gains and losses take into account only the effects of the trade and financial channels on wealth generated by the industrial sector. 56 18 Figure 4. Percent IPIARG variance due to: 25% 20% 15% 10% 5% 0% 1 3 5 7 9 11 Financial Channel 13 15 17 19 21 23 Trade Channel While the trade channel can never explain more than 5% of total IPIARG variance, the financial channel explains almost 25% of this variance in the long run. The joint analysis of the impulse-response and the variance decomposition indicate that in the event of a positive shock to the economy of the United States, the negative impact of the increase in FEDRATE on IPIARG dominates the positive effect of the increase in IMPOUA. In terms of equation (18) the final result of dollarization on real volatility will now depend on the sign of the coefficient of correlation between the cycles of Argentina and the United States. Given the previous results, if such coefficient is positive, dollarization could reduce real volatility. On the contrary, if the correlation between the cycles is negative, dollarization will increase the magnitude of cyclical fluctuations. Sensitivity analysis To evaluate the robustness of the results from the VECM, we implemented an analysis of sensitivity in three stages. These stages consist on examining the variability of the results of the model when we modify: 1) The short run restrictions (the order in which the variables are introduced in the VECM). 2) The specification of the deterministic component of the long run restrictions (the error correction mechanism). 3) The sample used for the estimation of the model. In the first case, the analysis responds to the criticism of some authors such as Leamer (1985) and Bernanke (1986) to the semi-automatic identification mechanism known as Cholesky decomposition. The criticism relies on the fact that the inferior triangular matrix that results from Cholesky’s decomposition is unique only until it is pre-multiplied by an orthogonal matrix (Hamilton, 1994). This implies that there could exist up to n! (the number of possible permutations if the number of variables in the model is n) different results depending on the order in which the variables are included in the model. If the covariance matrix of innovations is diagonal (that is, if shocks are not correlated), the ordering of the variables is irrelevant. In the opposite case, as indicated by Canova (1995a) and Enders (1995), one should check the sensitivity of the results estimating the impulse-response functions and the variance decomposition for every possible ordering, evaluating the dispersion of the results. The second stage of the analysis has two parts. First we check once again, through Johansen’s trace test, the number of cointegrating vectors for different specifications of the deterministic component. Secondly, we re-estimate the VECM with the different specifications for the deterministic component and evaluate the variability of the results of the impulse-response and variance decomposition of IPIARG. In the last stage of the analysis of sensitivity we evaluate the temporal stability of the results using a rolling 57 procedure similar to the one developed by Banerjee, Lumsdaine and Stock (1992) to analyze the shifting 57 Unlike recursive estimation (developed by Brown, Durbin and Evans,1975), the rolling coefficients are computed using sub-samples 19 root hypothesis. As a whole, the sensitivity analysis (presented in section 7 of the appendix) indicates that although there is a certain variability in the results (especially, the temporal instability of the trade channel effect on IPIARG from 58 the impulse-response function ), we can conclude that the results from the original VEC for the effect and relative importance of the different channels are strong and consistent. 8.2. Correlation between the cycles The second element to be taken into account to determine why it is convenient for Argentina to undertake a process of dollarization is, according to the analytical framework we have presented in the previous sections, the association between the business cycle of Argentina and the cycle of the United States. According to our findings thus far (the financial channel, which affects negatively IPIARG, is relatively more important than the commercial one, which affects it positively) if the cycles are synchronized (positively correlated) dollarization would reduce the volatility of Argentina’s business cycle (real volatility), reducing the country risk. This would be leaning the balance towards a position where the benefits of leaving the Peso would outweigh its costs (see equations 20 and 25). Even though the discussion on how to empirically calculate the business cycle is not part of this paper, it is of outmost importance to make a decision as regards the most adequate way to estimate it with the objective of studying the correlation. There are numerous ways of empirically calculating the cyclical component of a series but the literature has not found a definition that is usually most adequate (Canova, 1995b; Crivari-Neto, 1993). Given this theoretical disagreement we opt for presenting 4 different alternatives: a) The cycle from the first difference of the series, which assumes that the best specification for the behavior of the series is a random walk. This alternative is certainly better for the case of IPIUSA, which is I (1) meaning that differentiation eliminates completely the trend component of the series. In 59 the case of IPIARG, the evidence is somewhat contradictory . b) The cycle from the de-trending through the Hodrick-Prescott filter (Hodrick and Prescott, 1980), which implies a stochastic trend, independent of the behavior of the cyclical component. This mechanism of de-trending is widely used in the contemporaneous literature on the study of the business cycle (Backus and Kekoe, 1992; Carrera, Féliz and Panigo, 1998). c) The cycle assuming a linear trend, that is assuming a deterministic trend. d) The cycle from a linear trend with an endogenous structural break (using Perron’s 1994 test for the detection of the structural break) In table 3 we present the contemporaneous coefficients of correlation of the cycles of IPIARG and IPIUSA according to the 4 different specifications. that are a constant fraction δ0 of the full sample. In this way we can keep constant the marginal weight of each observation. 58 See panel b) in figure 11 in section 7 of the appendix. 59 We have discussed this problem in section 9.1 20 Table 3 Argentina’s cycle - USA’s cycle Contemporaneous correlation coefficients Industrial production index (USA) Industrial production index (ARGENTINA) Linear trend with First HodrickLinear trend Endogenous difference Prescott filter Structural Break* First difference 0.088 0.035 0.018 0.030 Linear trend 0.043 0.304 0.473 0.375 Linear trend with Endogenous Structural Break* -0.017 0.538 0.568 0.453 Hodrick-Prescott filter -0.142 0.478 0.334 0.329 Overall mean 0.24 Specific mean 0.32 * For Argentina the structural break was detected in 95:2. For the United States the break appeared in 93:8 In general, we find that for the period under analysis (1991:4 1999:10) the contemporaneous correlation between the business cycles of these countries is important and positive. The average value of the coefficient of contemporaneous correlation for the different specifications of the cycle is 0.24 (overall mean). Meanwhile, if we only take into account the combinations that belong to the principal diagonal on table 3 (that is, we analyze the correlation coefficients of the cycle calculated with the same specification for both countries) the average coefficient is even higher, 0.32 (specific mean). It is necessary to take into account that the coefficients from table 3 result from the analysis of the data of the complete sample period. However, it is possible that the correlation between the cycle is not constant in time. If the correlation were unstable the results should be taken with care. To evaluate the temporal evolution of the correlation between the business cycles we calculated the coefficients of correlation with a “rolling” methodology. Keeping the sample size fixed in 36 months (3 years), we calculated the coefficients of correlation changing the first and last element of the sample successively. Thus, the "rolling" coefficient of correlation associated with 94:3 includes data from the period 91:4-94:3, that 60 associated with 94:4 includes data from 91:5-94:4, and so on . Keeping the sample size fixed allows us to avoid the reduction in the weight of the marginal data point. We calculated the “rolling” coefficients for all crossed-correlations for − 3 ≤ i ≤ 3 , that is for the contemporaneous period and for the combinations that go from having IPIUSA lead in 3 months IPIARG to having IPIUSA lag in 3 months IPIARG. The results are presented as the average of the series calculated for each specification of the cycle. We analyze only two specifications: a) the first differences and c) assuming a linear trend. This allows us to establish extreme bands for the estimation of the cyclical correlation. The first one resulting from assuming a completely random trend in the series, and the other assuming, on the contrary, a completely deterministic trend for the series. These two specifications are presented in Figure 5. 60 We begin the "rolling" estimation in 94:3 since this is the first estimation that can be made including 36 months from the sample period under consideration (91:4 99:10). 21 Figure 5 Rolling crossed correlation coefficients IPIARG - IPIUSA 1.00 0.80 0.60 0.40 0.20 0.00 -0.20 -0.40 -0.60 Mar-94 Mar-95 Mar-96 Mar-97 1st difference Mar-98 Mar-99 Linear trend Note: The horizontal lines indicate the upper and lower limits of significativity of the coefficients at a 95% confidence level. While the “rolling” coefficient based on the first difference shows a relatively stable behavior in the period and very close to 0, the estimations based on assuming a linear trend show extreme volatility, although on average the coefficients are positive. This would confirm the previous results that indicate the existence of a significantly positive contemporaneous correlation between the business cycles of Argentina and that of the United States. The extreme volatility showed by the “rolling” coefficients in the short period under analysis raises serious doubts with respect to the usefulness of the most common specifications of the cycle (linear trend, for example) to define, with a certain degree of certainty, the question relating to Argentina’s convenience to leave its domestic currency in favor of the dollar. 9. Conclusions This study provides a non-traditional analytical framework for the evaluation of the convenience of an asymmetric monetary union (such as the dollarization) between a leading country and another that imports its monetary policy from it, from the point of view of the country that resigns its monetary sovereignty. The main objective of this paper is to evaluate how dollarization affects the real volatility of the country associating with the United States, and thus, how the loss of the exchange rate instrument impacts on the country risk (which depends positively on real volatility). With that in mind, we analyze the characteristics of the business cycles in both countries (the leader and the associated country) and the behavior of the trade and financial channels of transmission of shocks from the United States. In the literature there are two complementary visions to evaluate the convenience of a monetary union. The theory of the optimal currency areas (OCA) states that under conditions of symmetry of shocks, factor mobility, and price flexibility (amongst other conditions) it is efficient to resign the exchange rate instrument. The complementary vision relies on a financial analysis, and highlights that a monetary union is efficient when the cost of losing an instrument for covering against risk (such as the possibility of making use of exchange rate policy) is smaller than the gains derived from reduced exchange rate volatility resulting from discretionary policies. In this paper we found a more general expression for the analysis of dollarization from a financial point of view: "Dollarization will improve social welfare if the weight given by the policymaker (society) to the reduction in the aggregate risk (devaluation risk plus country risk) is greater than the loss of social welfare due to the reduction in the number of available instruments to cover for risk". The main contribution of this paper is to include in the analysis the effect of dollarization on country risk, through the impact of this asymmetric monetary union on the real volatility of the economy. This relationship 22 will depend on two central aspects: 1) the degree of synchronization between the cycles of the leader country and the associated country (AC), and 2) the effect and relative importance of the different channels of transmission of shocks from the leader to AC. 61 Under usual assumption regarding the way of transmission of the business cycle , the loss of monetary sovereignty of a country associating to the dollar will reduce real volatility if one of the following conditions is fulfilled: a) That the business cycles of the United States and of the associated country are positively correlated and that the financial channel is the main channel of transmission of the shocks that have origin in the United States, or b) That the business cycles are negatively correlated and that the trade channel dominates the financial channel. In both cases the clue relies on that the dollarization of the economy implies importing from the United States a counter-cyclical policy that is adequate to reduce real volatility (one of the main determinants of country risk). If, on the contrary, the cycles are negatively correlated and the financial channel dominates the trade channel, dollarization implies adopting a monetary policy that increases real volatility, augmenting the probability of default and thus country risk. In such circumstances, it is advisable to keep a national currency that allows us to use (albeit partially) a policy of automatic stabilization of the business cycle. The same recommendation applies to the case where the correlation between the cycles is positive but the trade channel dominates the financial one. In the empirical section of the paper we apply the model to evaluate the case of Argentina in the 1990s. The signs and magnitudes necessary were estimated using a number of econometric tools that included several cyclical correlation coefficients and the development of a vector error correction model (VECM). The main results are: 1) The financial channel has a negative impact on Argentina’s output. 2) The trade channel impacts positively on Argentina’s output. 3) The financial channel dominates the trade channel (the financial channel is the main means of transmission of shocks from the United States). 4) These results are robust to several stability tests for the VECM: changes in short run restrictions, different specifications of the deterministic component in the long run relationships and change in the sample of estimation. 5) The cycles of Argentina and the United States are on average positively correlated. 6) The correlation between the cycles is unstable through time and through different de-trending methodologies. With these results we can state that dollarization in Argentina would not only reduce the devaluation risk but also, it could reduce the real volatility of the economy and consequently the country risk. However, such strong result in the case of Argentina has to be interpreted carefully. The decision whether or not to dolarize an economy has to take into account the social value attached to the loss of an instrument of 62 risk diversification such as the exchange rate policy . Furthermore, Argentinean policymakers should not take such an important decision without considering 63 other alternatives that could even be more convenient . This issue requires, however, deeper discussion, which is in the line of our future research. A list of pending topics for the analysis of an asymmetric monetary union should include at least the following 61 That is, a positive shock in the economy of the United States affects positively AC’s GDP through the trade channel (through bigger imports for the United States) and negatively through the financial channel (due to the Federal Reserve’s counter-cyclical policy). These hypotheses are analyzed and corroborated empirically in the paper. 62 Within the Convertibility (a currency board) the instrument of risk edging is implicit in the possibility of leaving the monetary rule (temporarily or definitely) (i.e., the traditional restoration rule of Gold Standard, that was the base of the controversy between Winston Churchill and John Maynard Keynes. See on this Mc Kinnon, 1996). 63 For example integrating into a monetary union with Brazil (Argentina's main trading partner). In this alternative, Argentina would be trading off the benefits of avoiding the extreme bilateral exchange rate variability63 and the possibility of establishing a more leveled, symmetric union against the costs of importing Brazil's stronger domestic shocks 23 lines of research: 1) The long run correlation of the shocks that affect the members of the union, 2) the distribution of shocks and its relative size, 3) the real exchange rate volatility in the leader country and in those countries associating to it, 4) the influence of currency substitution on the decision to form an asymmetric monetary union, 5) the determinants of the productivity evolution and the effects of the association on long run convergence, 6) comparative advantages in an asymmetric monetary union and its impact on factor reallocation, and 7) the effects of an asymmetric monetary union on the labor market. The decision on dollarization implies a long run compromise from which it is almost impossible to retreat. There is no coming back from dollarization. For that reason, other factors (as the previous ones) should be taken into account when taking such a decision 10. References 1. Akaike, H. (1973), Information Theory and an Extension of the Maximum Likelihood Principle. In B. Petrov and F. Csake (eds), 2nd International Symposium on Information Theory, Budapest, Akademiai Kiado. 2. Amisano, G and Giannini, C. (1997), Topics in Structural VAR Econometrics. Springer-Verlag Berlin Heilderberg, New York. 3. Avila, J. (1998), Riesgo argentino y ciclo Económico. Universidad del Cema. Manuscript. 4. Backus, D.K. y Kekoe, P.J. (1992), International evidence on the historical properties of business cycles. American Economic Review, N°4. 5. Balino, T., Bennett, A. and Borensztein, E. (1999), Monetary Policy in Dollarized Economies, IMF, Occasional Paper N°. 171. 6. Bayoumi, T. and Eichengreen, B. (1992), Monetary and Exchange Rate Arrangements for NAFTA. NBER, Cambridge, Massachusetts, Manuscript. 7. Banerjee, A., Lumsdaine, R.L. and Stock, J.H. (1992), Recursive and sequential tests of the Unit Root and Trend Break hypothesis: theory and international evidence, Journal of Business and Economic Statistics, 10, 271-287. 8. Bergsten, F. (1999), Dollarization in emerging market economies and its policy implications for the United States. Institute for International Economics. (http://207.238.152.36/TESTMONY/dollariz.htm). 9. Bernanke, B. (1986), Alternative explorations of the money-income correlation, Carnegie Rochester Conference in Public Policy, 25, 49-100. 10. Bini Smaghi, L. and Vori, S. (1993), Ranking the EC as an Optimal Currency Area. Temi di Discussione, N°187, Banca d'Italia, Roma. 11. Blackburn, K. and Ravn, M. (1991), Contemporary macroeconomic fluctuations: an international perspective. . Working Paper 9106, University of Southampton. 12. Blanchard, O. and Quah, D. (1989), The Dynamic Effects of Aggregate Demand and Supply Disturbances. American Economic Review, 79, 655-673. 13. Brown, R.L., Durbin, J. and Evans, J.M. (1975), Techniques for testing the constancy of regression relationships over time. Journal of the Royal Statistical Society, Series B 37, 149-163. 14. Bootle, R. (1995), The economics of European Monetary Union. HSBC Markets. Manuscript. 15. Calvo, G. (1998), Capitals flows and capital-market crises: the simple economics of sudden stops. Journal of aplied Economics, Vol 1., N°1, November. 16. Calvo, G. and Végh, C. (1993), Exchange-rate based stabilisation under impefect credibility. In Frisch, H. and Worgotter, A., Open-economy macroeconomics. MacMillan Press. 17. Calvo, G., Leiderman, L. and Reinhart, C. (1993), Capital inflows and real exchange rate appreciation in Latin America: the rol of the external factors. IMF, Staff Papers. 18. Canova, F. and Ubide, A. J. (1997), International Business Cycles, Financial Markets and Household Production. Working Paper, 204, Discussion Papers Series, Universitat Pompeu Fabra. 19. Canova, F. (1995a), Vector autoregressive models: Specification, estimation, inference and forecasting. In Pesaran, M. and Wickens, M. (eds.), Handbook of applied econometrics, Blackwell P.L. 24 20. Canova, F . (1995b), Does Detrending Matter For The Determination Of The Reference Cycle And The Selection Of Turning Points? Economics Working Paper 113, Universitat Pompeu Favra, Universita di Catania, CEPR. 21. Canzoneri, M. B., Henderson, D. W. and Sweeney, R. (1987), Noncooperative monetary policies in interdependent economies: time consistency and reputation. Fall Academic Conference, Federal Reserve Bank, San Francisco. 22. Carraro, C. and Giavazzi, F. (1990), Can International Policy Coordination Really be Counterproductive?. In Carraro, C. (ed). International Economic Policy Coordination. Basil Blackwell Ltd., Oxford. 23. Carrera, J. (1995), Efecto Precio y Comercio en un Área Monetaria Asimétrica. Revista Económica, Año XLI, N°2, Octubre-diciembre. 24. Carrera J., Féliz M. and Panigo D. (1998), Análisis Integral de las Fluctuaciones Macroeconómicas en Argentina y Brasil. Documento Técnico CACES, N°1, CACES-UBA. 25. Carrera J., Féliz M. and Panigo D. (1999), Una medición de los canales de transmisión de las fluctuaciones económicas. El caso de Argentina y los Estados Unidos. Anales de la Asociación Argentina de Economía Política (AAEP), Rosario, Argentina. 26. Carrera, J.; Perez, P. and Saller, G. (1999), Currency Board and Tax Revenues, LACEA. 27. Cassella, A. (1993), Other People’s, Money: The Microfundation of Optimal Currency Areas. CEPR, Discussion Paper N°757. 28. Cohen, D. and Wyploz, C. (1990), France and Germany in the EMS: the Exchange Rate Constraint. In Alogoskoufis, G., Papademos, L. and Portes (eds)., Real External Constrains on Macroeconomic Policy: the European Experience. CPU, Cambridge. 29. Cooper, R. (1991), What Future for the International Monetary System? In Frenkel, J. and Goldstein, M. (eds)., International Financial Policy: Essays in Honor of Jacques Polak . IMF, Washington. 30. Chamie, N., DeSerres, A. and Lalonde, R. (1994), Optimum Currency Areas and Shock Asymetry: A Comparison of Europe and the United States. International Department, Bank of Canada, Ottawa. 31. Charemza, W. and Deadman D. (1997), New Direction In Econometric Practice. Elgar Pub. Lim. 32. Cribari Neto, F.(1993), The cyclical component in Brazilian GDP. Revista de Econometría, 1, 1-22. 33. De Grawe, P. (1988), Exchange Rate Variability and the Slowdown in Growth of International Trade. Staff Papers N°35, March. 34. De Grawe, P. (1992). The economics of monetary integration. Oxford University Press. 35. Dickey, D. and Fuller, W. (1979) Distribution of the estimators for a autorregresive time series with a unit root. Journal of the American Statistical Association, 74, 427-31. 36. Dooley, M. (1995), A Survey of Academic Literature on Controls over Capital Transactions. IMF, Working Paper 95/127. 37. Edwards, S. (1999), Interest rate volatility, contagion and convergence: an empirical investigation of the cases of Argentina, Chile and Mexico. Journal of Applied Economics, Vol 1, N°1, November. 38. Enders, W. (1995), Applied Econometric Time Series. J. Wiley and Song. 39. Fanelli. J.M. and González Rozada, M. (1998), Convertibilidad, Volatilidad y Estabilidad Macroeconómica en Argentina. Estudios de Política Económica y Finanzas, Revista de la Universidad de Palermo, Año 1, N°2, Octubre. 40. Fenton, P. and Murray, J. (1993), Optimum Currency Areas: A Cautionary tale. In The Exchange rate and the Economy. Bank of Canada. 41. Frenkel, R. (1998), Capital Market Liberalization and Economic Performance. Estudios de Política Económica y Finanzas, Revista de la Universidad de Palermo, Año 1, N°2, Octubre. 42. Ghosh, A. R. and Wolf, H. (1994), How many monies? A genetic approach to currency areas. Working Paper N°4805, NBER. 43. Giavazzi, F. and Giovannini, A. (1989), Limiting Exchange Rate Flexibility: The European Monetary 25 System. MIT Press, Cambridge, Massachusetts. 44. Hamilton, J. (1994), Time Series Analysis. Princeton University Press. 45. Helpman, E. and Razin, A. (1982), A comparison of exchange rate regimes in the presence of imperfect capital markets. International Economic Review, 23 (2), 365-88, June. 46. Hodrick, R. and Prescott, E. (1980), Poswar US business cycles:an empirical investigation. Discussion Paper No.441, Carnegie-Mellon University. 47. Horvart, J., Kandil, M. and Sharma, S. (1996), On the european Monetary System: the spillover Effects of German Shocks and desinflation. Department of Economics, University of Wisconsin- Milwaukee. Manuscript. 48. IDB (1995), Overcoming Volatility. Economic and Social Progress in Latin America. 1995 Report. 49. Johansen, S. (1988), Statistical analysis of cointegration vectors. Journal of Economic dynamic and control, 12, 231-54. 50. Johansen, S. (1991), Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models. Econometrica, 59, 1551–1580. 51. Johansen, S. (1995), Likelihood-based Inference in Cointegrated Vector Autoregressive Models, Oxford University Press. 52. Johansen, S. and Juselius, K. (1992), Testing structural hypotheses in a Multivariate cointegration analysis of the PPP and UIIP of UK. Journal of Econometric, 53, 211-44. 53. Kaldor, N. (1939), Speculation and Economic Stability. Review of Economics Studies, N°7. 54. Kenen, P. (1969), The Theory of Optimum Currency Areas: An Eclectic View. In Mundell, R. and Sowoboda, A. (eds), Monetary Problems of the International Economy. UCP. 55. Krugman, P. (1991), Policy Problems in a Monetary Union. MIT Press, Cambridge, Massachusetts. 56. Leamer, E. (1985), Vector autoregression for causal inference?. Carnegie Rochester Conference Series on Public Policy, 22, 255-304. 57. Levy Yeyati, E. and Sturzenegger, F. (1999), The Euro and Latin America. Is EMU a Blueprint for Mercosur? Business School, Universidad Torcuato Di Tella. Manuscript. 58. Masson and Taylor (1991), Common Currency Areas and currency unions: An Analysis of the Issues. DP N°617, CEPR. 59. McKinnon, R. I. (1963), Optimal Currency Areas. American Economic Review, Vol 52. 60. McKinnon, R. I. (1996), The rules of the game: international money and exchange rates. MIT Press, Cambridge, Massachusetts. 61. Mélitz, J. (1991), A Suggested reformulationa of the Theory of Optimal Currency Areas. DP N°590, CEPR. 62. Mundel, R. A. (1961), A Theory of Optimum Currency Areas. American Economic Review. Vol 51. 63. Neumeyer, P. A. (1998). Currencies and the allocation of risk: The welfare effects of a monetary union. The American Economic Review, Vol.88, N°1, 246-259. 64. Newey, W.K. and West, K.D. (1994), Automatic lags selection in covariance-matrix estimation. Review of Economics Studies, 61. 65. Perron, P. (1994), Trend, Unit Root and structural change in macroeconomic time series, in B.B.Rao (ed), Cointegration for the Applied Economist. Basingstoke, Macmillan Press. 66. Pesaran, M.H., Shin, Y. and Smith, R. (1999), Bounds testing approaches to the analysis of long run relationships. (http://www.econ.cam.ac.uk/faculty/pesaran/public.htm). 67. Phillips, P.C.B. and Perron, P. (1988), Testing for a Unit Root in time series regression. Biometrika, 75, 335-346, 1988. 68. Poole, W. (1970), Optimal Choice of Monetary Policy Instruments in a Simple Stocastich Macro Model. Quarterly Journal of Economics, N°85. 69. Posen, A. (2000), Dollarization, Currency Blocs, and US Policy. Institute for International Economics, 26 Washington. 70. Pou, P.(1999) Más dolarización para profundizar la convertibilidad, Diario Clarín 05/26/1999. 71. Ramos González, W., Vindas Garita, W. and Méndez Quesada, E. (1999), Viabilidad y conveniencia de la dolarización plena en Costa Rica. Departamento de Investigaciones Económicas, Banco Central de Costa Rica. Manuscript. 72. Rodríguez, C.A (1999), Macroeconomics Policies: Can We Transfer Lessons Across LDC’s. XII World Congress of International Economic Association, Buenos Aires. Manuscript. 73. Rogoff, K. (1991), Nachos and ECUs: On the Relative Merits of a North American Currency Union versus a European Monetary Union. University of California at Berkeley, Manuscript. 74. Roubini, N. (1999), An introduction to open macroeconomics, currency crises and the asian crisis. (http://www.stern.nyu.edu/~nroubini/asia/AsiaHomepage.html). 75. Roubistein, G. (1999), Dolarización. Argentina en la aldea global. Grupo Editor Latinoamericano, Buenos Aires. 76. Sachs, J. and Sala-I-Martin, X. (1991), Federal Fiscal Policy and Optimum Currency Areas. NBER working paper Nº. W3855, Cambridge, Massachusetts. 77. Schwarz, G. (1978), Estimating the dimension of a model. Annals of Statistics, 6, 461-4. 78. Schmitt-Grohé, S. (1998), The international transmission of economic fluctuations: Effects of US business cycle on the Canadian economy. Journal of International Economics, 44, 257-287. 79. Sims, C. (1980), Macroeconomics and Reality. Econométrica, 48, 1-49. 27 11. Appendix 11.1. The data: variables, source, frequency and sample size The empirical analysis takes as period of reference from April 1991 (91:4) to October 1999 (99:10), the last month for which data were available, with monthly periodicity, for every series used in the paper. The frequency as well as the sample period relate to the existence of a structural break at the beginning of the 90s in Argentina. This structural break has profoundly changed the behavior of the main macroeconomic variables of its economy. The implementation of the Convertibility Plan (in April, 1991) resulted in an evident reduction in the inflation rate, with a pronounced fall in the volatility of all nominal variables and an important increase in GDP, aggregate investment and consumption (Carrera, Féliz and Panigo, 1998). These profound changes in the cyclical and trend behavior of the Argentine variables prevented us from using longer samples, which would have included at the same time the pre and post Convertibility period. Even though a dummy variable could have taken into account the change in the trend, the alteration in the cyclical behavior of the series would result in reduced confidence in the results from models estimated in that sample. Under such circumstance we opted for using a sample period that begins with the Convertibility (April 1991) and ends in the last month were data was available for every series (October 1999). To avoid inefficient estimations, we needed enough degrees of freedom. For that reason, we chose to use monthly data which gave us 103 available observation for the period under study. As a proxy for the financial channel, we use the Federal Funds Rate of the Federal Reserve of the United 64 States (FEDRATE) (using the monthly average of daily data) since amongst the available rates this is the one that best adapts to our analytical framework where the financial channel transmits the monetary policy shocks generated by the Federal Reserve when it uses the interest rate as an automatic stabilizer of the United States’ economy. The trade channel is proxied by imports of the United States from Argentina (IMPOUA) (or, symmetrically, 65 exports from Argentina to the United States ). 66 As an approximation of the output we have used the Index of Industrial Production (IPI) for Argentina as 67 well as in the United States . We took as a proxy of the cycle of reference the IPI instead of the GDP because there are no trustworthy estimations of this last variable on a monthly periodicity. Anyway, both 68 variables (IPI and GDP) present high correlation in levels as well as cycles , for both countries when using 69 quarterly data . All the original series have been transformed through the application of the natural logarithm function. Afterwards, the series have been seasonally adjusted with the X-11 ARIMA procedure. The series for the real volatility of Argentina’s economy (ROLL_SD) results from estimating the standard deviation (SD) of the cycle of the Argentina’s IPI (IPIARG) in “rolling” sub-samples of 20 data points. Lastly, the series for Argentina’s country risk (CRISK) used results from calculating the difference between the implicit rate on Argentina’s government external bonds (Bonex 1989), denominated in United States 70 dollars, and 6 month Libor rate . For the analysis of the correlation of the business cycles in both countries we use the series IPIARG and IPIUSA, while for the construction of the VECM we work with the series IPIARG, EXPOAU and FEDRATE. The different econometric procedures used in this paper were developed under RATS 4.3 for Windows. 64 The data belong to the Federal Reserve Board’s database (FRED database). The data come from the database of the U.S. Census Bureau, Foreign Trade Division. 66 The data come from the Economy Ministry of Argentina. 67 The data belong to the Federal Reserve Board’s database (FRED database). 68 The cycle calculated with the Hodrick-Prescott filter for de-tending. 69 The coefficient of correlation between the cycle of Argentina’s IPI and the cycle of Argentina’s GDP (with quarterly data) are 0.865, while for the case of the United States the correlation between the cycle of the IPI and the cycle of the GDP (with quarterly data) is 0.876. On the other hand, the correlation between Argentina’s IPI and Argentina’s GDP (in levels, with quarterly data) is 0.840, whereas in the case of the United States the same correlation is of 0.991. 70 In both cases, the source of the data is the Central Bank of Argentina (BCRA). 65 28 11.2. Relationship between real volatility and country risk (Argentina) Given the contradictory theoretical positions on the subject, we proceeded to evaluate in the case of Argentina the relationship between real volatility of output (ROLL_SD) and country risk (CRISK). First, we estimated the coefficients of correlation between the variables (Table 4). Additionally, to check the stability of these coefficients we have reduced the sample sequentially eliminating the oldest periods to evaluate if the 1995 the Tequila crisis brought about any change in the correlation. While the correlation coefficient for the complete sample is 0.44 (contemporaneously), when we remove the data from older periods we find that the correlation not only gets stronger but that ROLL_SD begins to lead CRISK. 71 Then, we perform a linear regression of Argentina’s country risk (CRISK) as a function of real volatility of output (ROLL_SD). As in the case of the crossed correlation coefficients, real volatility has a strong, positive effect on the country risk (Table 5). In fact, an increase of 10% in real volatility produces a 2.2% increase in the country risk. Table 4 Cross correlation coefficient between IPIARG volatility (ROLL_SD) and Country risk (CRISK) ROLL_SD(t), CRISK (t+i) 91:4 - 99:10 t value 91:10 - 99:10 t value 92:4 - 99:10 t value 92:10 - 99:10 t value 93:4 - 99:10 t value 93:10 - 99:10 t value 94:4 - 99:10 t value 94:10 - 99:10 t value i=-6 i=-5 i=-4 i=-3 i=-2 i=-1 i=0 i=1 i=2 i=3 i=4 i=5 0.194 (1.92) 0.288 (2.83) 0.308 (2.95) 0.330 (3.07) 0.348 (3.13) 0.337 (2.89) 0.345 (2.82) 0.251 (1.89) 0.238 (2.40) 0.325 (3.26) 0.367 (3.61) 0.382 (3.64) 0.422 (3.95) 0.415 (3.71) 0.421 (3.60) 0.366 (2.89) 0.297 (3.07) 0.377 (3.89) 0.443 (4.55) 0.451 (4.49) 0.514 (5.11) 0.511 (4.86) 0.514 (4.68) 0.492 (4.19) 0.344 (3.62 0.411 (4.33) 0.495 (5.29) 0.506 (5.25) 0.592 (6.32) 0.582 (5.90) 0.577 (5.56) 0.584 (5.38) 0.389 (4.20) 0.415 (4.39) 0.514 (5.59) 0.526 (5.56) 0.612 (6.71) 0.613 (6.44) 0.609 (6.09) 0.625 (6.05) 0.427 (4.72) 0.421 (4.50) 0.529 (5.85) 0.549 (5.95) 0.628 (7.03) 0.639 (6.94) 0.636 (6.60) 0.654 (6.58) 0.440 (4.93) 0.418 (4.49) 0.529 (5.88) 0.561 (6.17) 0.628 (7.08) 0.644 (7.09) 0.641 (6.73) 0.657 (6.70) 0.433 (4.81) 0.421 (4.50) 0.529 (5.85) 0.564 (6.18) 0.633 (7.12) 0.651 (7.18) 0.646 (6.76) 0.663 (6.74) 0.422 (4.63) 0.420 (4.47) 0.530 (5.82) 0.566 (6.18) 0.638 (7.18) 0.661 (7.32) 0.654 (6.85) 0.676 (6.92) 0.392 (4.21) 0.390 (4.06) 0.496 (5.30) 0.534 (5.66) 0.604 (6.52) 0.631 (6.71) 0.62 (6.23) 0.643 (6.28) 0.345 (3.62) 0.336 (3.40) 0.444 (4.56) 0.484 (4.92) 0.550 (5.63) 0.575 (5.76) 0.56 (5.28) 0.587 (5.38) 0.293 (3.00) 0.278 (2.75) 0.390 (3.89) 0.430 (4.21) 0.490 (4.77) 0.514 (4.87) 0.497 (4.44) 0.528 (4.56) Sample Size 0.247 103 (2.48) 0.228 97 (2.21) 0.345 91 (3.35) 0.387 85 (3.68) 0.440 79 (4.13) 0.457 73 (4.15) 0.441 67 (3.77) 0.478 61 (3.96) i=6 Table 5 Linear regression using "General to Specific Methodology": CRISK on ROLL_SD Dependent Variable: CRISK Method: Least Squares Sample(adjusted): 1992:09 1999:04 Included observations: 80 after adjusting endpoints Variable Coefficient Std. Error t-Statistic Prob. CRISK(-1) 0.56 0.07 7.59 0.00 CRISK(-3) 0.18 0.07 2.54 0.01 CRISK(-6) 0.15 0.08 1.98 0.05 CRISK(-9) -0.35 0.12 -2.99 0.00 CRISK(-10) 0.48 0.12 4.11 0.00 CRISK(-12) -0.54 0.09 -5.85 0.00 CRISK(-14) -0.50 0.14 -3.66 0.00 CRISK(-15) 0.88 0.12 7.25 0.00 ROLL_SD 0.22 0.10 2.10 0.04 ROLL_SD(-3) 0.29 0.12 2.47 0.02 ROLL_SD(-7) -0.77 0.21 -3.65 0.00 ROLL_SD(-8) 0.60 0.19 3.18 0.00 C 1.24 0.46 2.68 0.01 R2=0.904 DW=2.06 71 Using OLS for the estimation of the coefficients and the methodology “General to Specific” to obtain the most parsimonious form for the equation (see Charemza and Deadman, 1997). 29 11.3. Relationship between the variables of the hypothesis in equations (2), (3), and (5) Several papers have asserted the existence of a strong positive correlation between the international interest rate and the domestic interest rate of Latin American countries (Calvo, Leiderman and Reihart, 1993). In the specific case of Argentina, Carrera, Féliz and Panigo (1998) find that there is an strong correlation between the cycles of the Federal Reserves Funds Rate and Argentina's deposit rate (0.36, contemporaneously). 72 We perform a correlation analysis of the FEDRATE and IPIUSA , and we find that there exists a strong, positive correlation (0.60) between the variables (Table 6). For bilateral trade, we assume that an increase in the United States’ output results in an increase in the US’ imports from Argentina (IMPOUA). In Table 6 we show that the crossed correlation coefficient between IPIUSA and IMPOUA show that the correlation is positive and contemporaneous (0.87). Table 6 Cross-correlation coefficient between IPIUSA and the different channels of transmission i 12 11 10 9 8 7 6 5 4 3 2 1 0 -1 -2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 72 IMPOUA(t),IPIUSASA(t+i) 0.638 0.655 0.678 0.693 0.716 0.748 0.760 0.775 0.797 0.821 0.845 0.861 0.874 0.855 0.839 0.818 0.809 0.787 0.758 0.716 0.685 0.663 0.633 0.605 0.581 FEDRATESA(i),IPIUSASA(t+i) 0.461 0.466 0.471 0.475 0.480 0.485 0.489 0.492 0.495 0.501 0.508 0.516 0.525 0.542 0.555 0.567 0.579 0.588 0.595 0.599 0.598 0.592 0.580 0.568 0.553 In the present paper we use as a proxy of the United States’ output the Industrial Production Index for the motives expressed in 11.1 30 11.4. Unit Root Tests Table 7. ADF and Phillips Perron Tests for Unit Root Variable: IPIARG ADF test Test Structure Critical Value 5% Levels 1st Dif. Intercept -1.79 -6.20 -2.89 Intercept&Trend -2.28 -6.18 -3.46 None 0.77 -6.16 -1.94 PP test Test Structure Critical Value 5% Levels 1st Dif. Intercept -2.85 -17.09 -2.89 Intercept&Trend -4.14 -17.15 -3.45 None 0.92 -16.98 -1.94 Variable. FEDRATE ADF test Test Structure Critical Value 5% Levels 1st Dif. Intercept -1.91 -2.92 -2.89 Intercept&Trend -2.40 -2.92 -3.45 None -0.03 -2.98 -1.94 PP test Test Structure Critical Value 5% Levels 1st Dif. Intercept -1.47 -6.50 -2.89 Intercept&Trend -2.29 -6.69 -3.45 None -0.51 -6.52 -1.94 Variable: IPIUSA ADF test Test Structure Critical Value 5% Levels 1st Dif. Intercept 0.05 -6.09 -2.89 Intercept&Trend -1.83 -5.58 -3.45 None 6.00 -1.37 -1.94 PP test Test Structure Critical Value 5% Levels 1st Dif. Intercept -0.27 -15.38 -2.89 Intercept&Trend -2.64 -15.30 -3.45 None 9.72 -10.07 -1.94 Variable: IMPOUA ADF test Critical Value 5% Levels 1st Dif. -0.86 -13.62 -2.89 -3.13 -13.57 -3.45 0.85 -13.60 -1.94 PP test Critical Value 5% Levels 1st Dif. -2.42 -23.82 -2.89 -7.53 -23.73 -3.45 0.65 -23.55 -1.94 Test Structure Intercept Intercept&Trend None Test Structure Intercept Intercept&Trend None 31 11.5. Test to determine the optimal lag length for the VEC model Table 8. Information criteria Lag structure 10 lags 9 lags 8 lags 7 lags 6 lags 5 lags 4 lags 3 lags 2 lags 1 lag Akaike -8.85 -8.91 -8.98 -9.03 -9.00 -9.09 -9.24 -9.21 -8.98 -8.56 Schwarz -6.31 -6.62 -6.94 -7.23 -7.44 -7.78 -8.18 -8.40 -8.41 -8.24 Table 9. Likelihood Ration Test H0 Lag 10 is insignificant Lag 9 is insignificant Lag 8 is insignificant Lag 7 is insignificant Lag 6 is insignificant Lag 5 is insignificant Lag 4 is insignificant Lag 3 is insignificant Lag 2 is insignificant Likelihood ratio test Log of Log of Observed Significance Chi-Squared level ΩU ΩR -19.36 -19.23 8.22 0.51 -19.23 -19.11 7.93 0.54 -19.11 -18.96 9.90 0.36 -18.96 -18.74 16.04 0.07 -18.73 -18.64 7.37 0.60 -18.64 -18.59 3.53 0.94 -18.59 -18.37 17.47 0.04 -18.37 -17.95 35.00 0.00 -17.95 -17.33 53.10 0.00 Note: ΩU and ΩR represent the covariance matrix of the residuals of the unrestricted and restricted models, respectively. Table 10. P-values for Serial Correlation LM Test on the VEC model residual Lag length in VAR model VAR equation 10 0.54 FEDRATE (0.89) 0.50 2 lags IMPOUA (0.95) 0.70 IPIARG (0.72) 0.87 FEDRATE (0.53) 0.36 4 lags IMPOUA (1.11) 0.33 IPIARG (1.16) 0.89 FEDRATE (0.49) 0.84 7 lags IMPOUA (0.57) 0.64 IPIARG (0.79) Note: In parenthesis we present the values for the F-statistic. Lags in the Breusch-Godfrey equation 9 8 7 6 0.59 0.50 0.55 0.60 (0.83) (0.93) (0.85) (0.76) 0.91 0.85 0.79 0.97 (0.44) (0.50) (0.56) (0.23) 0.85 0.79 0.75 0.75 (0.52) (0.58) (0.61) (0.57) 1.00 1.00 0.99 0.98 (0.14) (0.16) (0.16) (0.17) 0.90 0.84 0.76 1.00 (0.45) (0.51) (0.59) (0.05) 0.86 0.84 0.76 0.71 (0.52) (0.52) (0.59) (0.62) 0.96 0.94 0.90 0.85 (0.33) (0.35) (0.40) (0.44) 1.00 1.00 1.00 0.99 (0.13) (0.13) (0.13) (0.13) 0.92 0.92 0.97 0.95 (0.42) (0.40) (0.27) (0.27) Average 0.56 0.80 0.77 0.97 0.77 0.70 0.91 0.96 0.88 Table 11. P-values for ARCH LM Test on the VEC model residual Lag length in VAR model VAR equation 10 0.58 FEDRATE (0.86) 0.83 2 lags IMPOUA (0.57) 0.42 IPIARG (1.04) 0.50 FEDRATE (0.94) 0.58 4 lags IMPOUA (0.85) 0.81 IPIARG (0.59) 0.25 FEDRATE (1.29) 0.25 7 lags IMPOUA (1.29) 0.71 IPIARG (0.72) Note: In parenthesis we present the values for the F-statistic. Lags in the Breusch-Godfrey equation 9 8 7 6 0.49 0.39 0.40 0.33 (0.95) (1.08) (1.05) (1.18) 0.81 0.70 0.84 0.68 (0.58) (0.69) (0.50) (0.66) 0.54 0.49 0.46 0.37 (0.88) (0.94) (0.97) (1.09) 0.65 0.55 0.50 0.44 (0.76) (0.87) (0.91) (0.99) 0.57 0.55 0.39 0.26 (0.86) (0.87) (1.06) (1.31) 0.78 0.69 0.62 0.55 (0.61) (0.71) (0.76) (0.83) 0.17 0.11 0.41 0.39 (1.49) (1.70) (1.04) (1.07) 0.20 0.13 0.09 0.05 (1.40) (1.62) (1.82) (2.19) 0.80 0.73 0.67 0.60 (0.59) (0.65) (0.70) (0.76) 32 Average 0.44 0.77 0.46 0.53 0.47 0.69 0.27 0.15 0.70 Table 12. Std. Deviation, Skewness and Kurtosis of the VEC model residuals Lag length in VEC model 2 lags 4 lags 7 lags 11.6. Statistic Std. Deviation Skewness Kurtosis Std. Deviation Skewness Kurtosis Std. Deviation Skewness Kurtosis FEDRATE 0.03 -0.59 5.93 0.03 -0.37 7.27 0.03 -0.47 5.34 VAR Equation IMPOUA 0.14 -0.01 4.13 0.14 -0.07 3.71 0.13 0.42 3.67 Unit Root test for the residual from error correction mechanism. Table 13. ADF and Phillips Perron Tests for Unit Root Variable: Residual from Error Correction Mechanism ADF test Test Structure Critical Value 5% Levels 1st Dif. Intercept -4.25 -3.63 -2.89 Intercept&Trend -4.16 -3.63 -3.46 None -4.30 -3.63 -1.94 PP test Test Structure Critical Value 5% Levels 1st Dif. Intercept -4.08 -16.31 -2.89 Intercept&Trend -4.13 -16.23 -3.45 None -4.11 -16.35 -1.94 33 IPIARG 0.03 -0.29 6.96 0.03 -0.70 6.21 0.04 -0.05 6.36 11.7. Sensitivity analysis results. A) Sensitivity of the results to the order in which the variables are included in the VECM A.1) Sensitivity of the impulse-response function. Figure 6. IPIARG average response to one standard innovation in: a) Financial Channel b)Trade Channel 0.5% 0.5% 0.0% 0.0% -0.5% -0.5% -1.0% -1.0% -1.5% -1.5% 1 4 7 10 13 16 19 22 1 4 7 10 13 16 19 22 Note: the line with symbols and the full line represent IPIARG (for the 6 possible orderings) average response to one standard innovation in the financial and trade channel, respectively. The dotted lines represent those average values +/- one standard deviation calculated from the different possible orderings for each variable. A.2) Sensitivity of the variance decomposition. Figure 7. Average percent IPIARG variance due to: a) Financial channel b) Trade Channel 30.0 30.0 25.0 25.0 20.0 20.0 15.0 15.0 10.0 10.0 5.0 5.0 0.0 0.0 1 4 7 10 13 16 19 22 1 4 7 10 13 16 19 22 Note: the line with symbols and the full line represent IPIARG (for the 6 possible orderings) average percent variance due to the financial and trade channel, respectively. The dotted lines represent those average values +/- one standard deviation calculated from the different possible orderings for each variable. 34 B) Sensitivity of the results to a change in the deterministic component used to estimate the error correction mechanism B.1) Sensitivity of Johansen Trace test. Table 14. Cointegration Analysis Trace Test for cointegrating vectors Structure of the H0 (number of cointegrating deterministic component vectors): Eigenvalue Likelihood Ratio 5% Critical value 1% Critical Value Alternative 1: No trend in None * data and intercept in the At most 1 error correction mechanism At most 2 0.179 36.22 34.91 41.07 0.122 15.86 19.96 24.6 0.024 2.50 9.24 12.97 Alternative 2: Linear trend None * in data and intercept in the error correction At most 1 mechanism At most 2 0.173 33.47 29.68 35.65 0.120 13.79 15.41 20.04 0.005 0.57 3.76 6.65 Alternative 3: Linear trend None ** in data and Linear trend At most 1 and intercept in the error correction mechanism At most 2 0.217 49.34 42.44 48.45 0.131 24.18 25.32 30.45 0.089 9.66 12.25 16.26 Note: *(**) denotes rejection of the hypothesis at 5%(1%) significance level B.2) Sensitivity of the impulse-response function. Figure 8. IPIARG response to one standard innovation in: a) Financial channel b) Trade Channel 0.6% 0.2% 0.5% 0.0% 0.4% -0.2% 0.3% -0.4% -0.6% 0.2% -0.8% 0.1% -1.0% 0.0% -1.2% -0.1% -1.4% -0.2% -1.6% -0.3% 1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23 Note: For panel a) as well as for panel b) from figure 8, the bold line represents results from alternative 1 in table 14, while the thin line and the dotted line represent alternative 2 and 3, respectively. 35 B.3) Sensitivity of the variance decomposition. Figure 9. Percent IPIARG variance due to: a) Financial channel b) Trade Channel 30.0 30.0 25.0 25.0 20.0 20.0 15.0 15.0 10.0 10.0 5.0 5.0 0.0 0.0 1 3 5 7 9 11 13 15 17 19 21 23 1 3 5 7 9 11 13 15 17 19 21 23 Note: For panel a) as well as for panel b) from figure 9, the bold line represents results from alternative 1 in table 14, while the thin line and the dotted line represent alternative 2 and 3, respectively. C) Sensitivity of the results to changes in the sample used to calculate the coefficients of the VECM 1 C.1) Sensitivity of the impulse-response function. Figure 10. IPIARG average response to one standard innovation in: a) Financial Channel b)Trade Channel 1.0% 2.0% 0.5% 1.5% 0.0% 1.0% -0.5% 0.5% -1.0% 0.0% -1.5% -0.5% -2.0% -1.0% 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 Note: The line with symbols and the full line represent IPIARG average response (of the 44 rolling sub-samples) to one standard innovation in the financial and trade channel, respectively. The dotted lines represent those average values +/- the standard deviation calculated from the 44 sub-samples analyzed. 1 For this analysis we take 44 "rolling" sub-samples with 60 observations each, beginning the estimation with the subsample that starts in 91:4 and ends in 96:3. The last sub-sample includes the period 94:11 99:10. 36 Figure 11. IPIARG Rolling average response to one standard innovation in: a) Financial Channel b)Trade Channel 0.50% 1.50% 0.00% 1.00% -0.50% 0.50% -1.00% 0.00% -1.50% -2.00% Mar-96 Sep-96 Mar-97 Sep-97 Mar-98 Sep-98 Mar-99 Sep-99 -0.50% Mar-96 Sep-96 Mar-97 Sep-97 Mar-98 Sep-98 Mar-99 Sep-99 Note 1: The line with symbols and the full line represent IPIARG rolling average (of the 48 first months after the shock) response to one standard innovation in the financial and trade channel, respectively. Note 2: Percent of the sub-samples with negative response of IPIARG to a shock of one standard innovation in the financial channel: 83%. Note 3: Percent of the sub-samples with positive response of IPIARG to a shock of one standard innovation in the trade channel: 70%. C.2) Sensitivity of the variance decomposition. Figure 12. Average percent IPIARG variance due to: 40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 Financial channel Trade channel Note: The line with symbols and the full line represent the average (of the 44 subsamples) percent IPIARG variance explained by the FEDRATE and IMPOUA, respectively. 37