Financial conditions and monetary policy in Uruguay: An MS-VAR approach

Bucacos Iguini, Elizabeth Working Paper Financial conditions and monetary policy in Uruguay: An MS-VAR approach IDB Working Paper Series, No. IDB-WP-796 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Bucacos Iguini, Elizabeth (2017) : Financial conditions and monetary policy in Uruguay: An MS-VAR approach, IDB Working Paper Series, No. IDB-WP-796, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0000699 This Version is available at: https://hdl.handle.net/10419/173863 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/3.0/igo/legalcode Financial Conditions and Monetary Policy in Uruguay: A n MS-VAR Approach Elizabeth Bucacos IDB WORKING PAPER SERIES Nº IDB-WP-796 A pril 2017 Department of Research and Chief Economist Inter-American Development Bank A pril 2017 Financial Conditions and Monetary Policy in Uruguay: A n MS-VAR Approach Elizabeth Bucacos Banco Central del Uruguay Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Bucacos, Elizabeth. Financial conditions and monetary policy in Uruguay: an MS-VAR approach / Elizabeth Bucacos. p. cm. — (IDB Working Paper Series ; 796) Includes bibliographic references. 1. Financial crises-Uruguay-Econometric models. 2. Monetary policy-UruguayEconometric models. 3. Uruguay-Economic conditions. I. Inter-American Development Bank. Department of Research and Chief Economist. II. Title. III. Series. IDB-WP-796 Copyright © Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 AttributionNonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/ legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this CC-IGO license. Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised version of this work may also be reproduced in any academic journal, including those indexed by the American Economic Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. Therefore, the restriction to receive income from such publication shall only extend to the publication's author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives license and these statements, the latter shall prevail. Note that link provided above includes additional terms and conditions of the license. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. http://www.iadb.org 2017 1 Abstract1 This study analyzes the effects of “financial stress” on the Uruguayan macroeconomy in the 1998Q3-2016Q2 period with the underlying idea that financial shocks propagate differently during “normal times” than during times of “stress.” This behavior is captured in a multivariate framework through a Markovswitching vector auto regressive (MS-VAR) model. The evidence found so far supports the idea that financial conditions affect the macroeconomy, as they not only change the private investment long-run average growth rate but also directly modify the behavior of monetary policy. JEL classifications: C34, E27, E44, E62 Keywords: Switching-regression models, Investment, Financial markets and the macroeconomy, Uruguay 1This research was carried out within the framework of CEMLA’s Joint Research Program 2016 coordinated by the Central Bank of Brazil. The author thanks counseling and technical advisory provided by the Financial Stability and Development Group of the Inter-American Development Bank in the process of writing this document. The opinions expressed in this publication are those of the author and do not reflect the views of CEMLA, the EDF group, the Inter-American Development Bank or the Central Bank of Uruguay. She wants to thank for their comments and suggestionsGerardo Licandro, Serafín Frache, Waldyr Dutra, Oscar Carvallo, Jaime Martínez, Roberto Chang, Pablo Guerrón, Andrés Fernández, Carolina Pagliacci and participants of CEMLA Joint Research 2016 at the Workshop in Mexico City and at the XXI Annual Meeting of the Central Bank Researchers Network in Brasilia. Contact information for the author: Elizabeth Bucacos, Banco Central del Uruguay, Diagonal Fabini 777, Montevideo, Uruguay; email: [email protected] 2 1. Introduction According to Modigliani-Miller (1958) theorem, the financial structure is both indeterminate and irrelevant to real economic outcomes. Nevertheless, the consensus now considers this assumption to be only a simplified tool in model designing and a good premise only when financial frictions are small. Bernanke, Gertler and Gilchrist (1998: 1343) wisely point out: … However, as Gertler (1988) discusses, there is a long-standing altertantive tradition in macroeconomics, beginning with Fisher and Keynes if not earlier authors, that gives a more central role to credit market conditions in the propagation of cyclical fluctuations. In this alternative view, deteriorating creditmarket conditions—sharp increases in insolvencies and bankruptcies, rising real debt burdens, collapsing asset prices, and bank failures—are not simply passive reflections of a declining real economy, but are in themselves a major factor depressing economic activity. There is a vast literature characterized by asymetric information and agency problems where the Modigliani-Miller irrelevance theorem no longer applies. For instance, credit-markets malfunctions may increase the real cost of new credit and reduce the efficiency of matching potencial borrowers and lenders, which may negatively affect real output and employment. As a result, nowadays it is widely accepted that there are important links between the financial sector and the macroeconomy. Work along this line includes, among others, Bernanke and Blinder (1988), Kashyap, Stein and Wilcox (1993), Kashyap and Stein (1994), Hubbard (1998), Bernanke, Gertler and Gilchrist (1999), and Céspedes, Chang and Velasco (2000). More recently, after the occurrence of the subprime mortgage crises in 2008, it has become evident that stress in financial markets may affect economic activity. Many central banks and financial agencies began to collect and analyze different kinds of information in the hope of understanding the phenomenon and creating tools with predictive value. Still, there is no consensus on how to define or measure financial stress. Hubrich and Tetlow (2014) point out that it is hard to find evidence of the link between the financial sector and the macroeconomy during normal times once monetary and other factors are accounted for. According to their view, one reason why statistically significant and 3 macroeconomically important linkages have been elusive is because the importance of financial factors has tended to be episodic in nature. They argue that financial frictions become more important when the financial system is not operating normally. They conclude that it seems reasonable to examine the interdependency of the financial sector and the macroeconomy in a nonlinear multivariate framework. They use a richly parameterized Markov-switching VAR (MS-VAR) model estimated using Bayesian methods. That approach seems appealing and appropriate for explaining the Uruguayan performance. Almost since its independence in the second half of the nineteenth century, Uruguay has experienced stop-and-go episodes closely related to an adverse financial event (banking crisis, exchange rate crisis, external debt crisis, etc.).2 Figure 1. Uruguayan Private Consumption Panel (a) Index Panel (b) Year differenced 2 For a more comprehensive analysis, see Barrán and Nahun (1967-1978) and Vaz (1999). 60 80 100 120 140 160 180 200 94 96 98 00 02 04 06 08 10 12 14 16 PRIVATE CONSUMPTION Source: Own calculations based on BCU data. Index, quarterly deseasonalized data Triplet crisis Subprime crisis Brazilian crisis Tequila effect -30 -20 -10 0 10 20 94 96 98 00 02 04 06 08 10 12 14 16 Triplet crisis Subprime crisis Tequila effect Brazilian crisis PRIVATE CONSUMPTION Year differenced Source: Own calculations based on BCU data. 4 Figure 1 depicts the evolution of Uruguayan private consumption. It shows a rising pattern for most of the 1993-2016 period, excepting some specific episodes which can be related to some kind of financial turbulence. We can identify a few main events. In 1994-5, the Tequila effect, a sudden devaluation in the Mexican peso, caused other currencies in the region (mainly Southern Cone and Brazil) to decline, leading to an income fall given the important level of dollarization of the Uruguayan economy.3 In 2001-2 there was a triple crisis in Uruguay4 (balance of payments, banking and fiscal crisis). Third were the domestic consequences of the 2008 subprime crisis. Finally, a drop toward the end of the sample was probably related to the deterioration of the political and economic situation in Brazil.5 Although they are only coincidences, they are indicative of some correlation between the financial markets and the macroeconomy. Figure 2 presents almost the same evidence for the evolution of private investment.6 Figure 2. Uruguayan Private Investment Panel (a) Index 3 Since both the Uruguayan Government and firms had high levels of U.S. dollar-denominated debt, the devaluation reduced disposable income and made it increasingly difficult to pay back tdebts. 4 A more detailed explanation is found in Section 3.2. 5 The influence of Brazil, one of Uruguay’s main trade partners, is reflected in the weight of Brazilian currency in the effective Uruguayan real exchange rate: 30 percent. 6 Measured as gross fixed capital formation by the private sector, including buildings and machinery and equipment. 86 88 90 92 94 96 98 100 102 104 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 Triplet crisis Subprime crisis Brazilian crisis PRIVATE INVESTMENT Index, quarterly deseasonalized data Source: Own calculations based on BCU data 5 Figure 2, continued Panel (b) Year differenced In this study, it is asssumed that stress events are episodic in nature. It is also assumed that their exact occurrence is unknown beforehand but that they have a non-negligible probability of appearance. There is an underlying idea that shocks propagate differently during “normal times” than during times of “stress,” which is captured in a nonlinear multivariate framework through a Markov-switching vector auto regressive (MS-VAR) model. Following Hubrich and Tetlow (2014), a “stress event” is defined as a period where the latent Markov states for both shock variances and model coefficients are adverse7. The rest of the paper is organized as follows. Section 2 explains and justifies the methodology used, presents the data and gives details of an estimated coincident FCI to be incorporated into the MS-VAR. Section 3 estimates the MS-VAR model and identifies the transmission mechanisms. Section 4 concludes. 2. Methodology This section deals with methodological issues. First of all, statistical indicators to measure financial stability are explained and a financial conditions index (FCI) is constructed. Next, Markov-switching models are analyzed. Then, an MS-VAR model for Uruguayan data is presented. 7 The present document does not allow for changing parameters; that approach is left for future research. -8 -6 -4 -2 0 2 4 6 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 PRIVATE INVESTMENT Year differenced Triplet crisis Brazilian crisis Subprime crisis Source: Own calculations based on BCU data. 12 Hamilton (1989) introduced a random variable St that represents the unobserved regime or state of the economy: (1) 𝑦𝑡=𝜈(𝑆𝑡)+𝜖𝑡, 𝜖𝑡 ~𝐼𝐼𝐼(0, 𝜎2) for 𝑆𝑡 ∈{0,1,2, … , 𝑆−1} The objective of his model is to estimate the probabilities of being in a regime, together with the other model parameters. The unobserved random variable St follows a Markov chain, defined by transition probabilities between the S states: (2) 𝑔𝑖 𝑗 ⁄= 𝑃[𝑆𝑡+1=𝑖/𝑆𝑡=𝑗], 𝑖,𝑗= 0, 1, … , 𝑆−1 The probability of being in a regime, given available data and past regimes, only depends on the previous regime (and available data); there is no benefit from knowing the whole history of the model. Then, the probabilities of moving from one regime to another (“transition probabilities”) have a Markovian structure: (3) 𝑃[𝑆𝑡+1 =𝑖 / 𝑆𝑡=𝑗,𝑆𝑡−1,𝑆𝑡−2, … ]=𝑃[𝑆𝑡+1 =𝑖/𝑆𝑡=𝑗], 𝑖,𝑗= 0, 1, … , 𝑆−1 Because the system has to be in one of the S states: (4) �𝑔𝑖/𝑗 𝑆−1 0= 1 the extension to the multivariate framework is straightforward.19 A conventional VAR(1) can be written as: (5) 𝑦𝑡=𝜈+𝜋1𝑦𝑡−1+𝜀𝑡, 𝜀𝑡 ~𝐼𝐼𝑛(0, Σ) where 𝑦𝑡,𝜈,𝜀𝑡 are nx1 vectors and 𝜋1,Σ are nxn matrices. The MS-VAR model allows for a great variety of specifications. Krolzig (1997) has established a unique notation for each model, adding to the general MS term the regimedependent parameters; if exogenous regressors are included into de system, it is denoted as MSVARX. See Table 2. 19 This closely follows Chapter 14 of Doornik (2013). 13 Table 2. MS-VAR Notation M Markov-switching mean MSM-VAR I Markov-switching intercept MSI-VAR A Markov-switching autoregressive parameters MSA-VAR H Markov-switching heteroskedasticity MSH-VAR Source: Krolzig (1997). For instance, a VAR(1) with Markov-switching in intercept and variances, MSIH(S)-VAR(1), can be written as: (6) 𝑦𝑡(𝑆𝑡)=𝜈(𝑆𝑡)+𝜋1𝑦𝑡−1+𝜀𝑡, 𝜀𝑡 ~𝐼𝐼𝑛�0, Σ(𝑆𝑡)� Within each S regime, the specification is a conventional VAR. The density of yt conditional on the state St = j, is a multivariate normal: (7) 𝑓(𝑦𝑡(𝑆𝑡)|𝑆𝑡)=[(2𝜋)𝑛|Σ(𝑆𝑡)|]−1 2 𝑒𝑥𝑔�−1 2𝑣𝑡(𝑆𝑡)′ Σ(𝑆𝑡)−1 𝑣𝑡(𝑆𝑡)� where 𝑣𝑡(𝑆𝑡)=𝑦𝑡(𝑆𝑡)−𝜈(𝑆𝑡)−𝜋1𝑦𝑡−1 and |Σ| is the determinant of Σ. Then, the procedure consists in maximizing the loglikelihood as usual. In order to solve the identification problem,20 a convenient way is to apply Choleski decomposition: (8) Σ=𝑃𝑃′ where P is lower diagonal with positive diagonal elements. This can be written as: (9) Σ=𝐵 𝑆2𝐵′ where B is lower diagonal with ones on the diagonal and S is diagonal. This has the advantage of making it easier to get the determinant and inverse of Σ. In this investigation three different specifications are tried, as presented in Table 3. Table 3. MS-VAR: Different types of variance Type of variance Specification Parameters in scale in B Fixed Σ= B S2B′ n n(n-1)/2 Switching scale Σ(𝑆𝑡)= BS2(𝑆𝑡)B′ Sn n(n-1)/2 Switching variance Σ(𝑆𝑡)= B(𝑆𝑡)S2(𝑆𝑡)B′(𝑆𝑡) Sn Sn(n-1)/2 Note: Σ is nxn variance-covariance matrix, B is lower diagonal with ones in the diagonal and S is diagonal; St denotes the unknown states or regimes. 20 The identification problem deals with the fact that �𝑛2−𝑛� 2 restrictions between the regression residuals and the structural shocks are required in order to recover deep parameters from the reduced-form parameters in a VAR. 14 2.4 Discussion A structural break is clearly recognized when there is an unexpected shift in a time series. When there are too many unknown breaks it is assumed that the parameters are time-varying. This is not the way the Uruguayan financial situation seems to be characterized in the time period analyzed here, because it shows a picture of relative stability interrupted by a few episodes of roughness (Figure 3). In addition, both private consumption (see Figure 1) and private investment (Figure 2) evolve in a similar way. In a globalized economy, financial frictions may appear without previous notice from different places, disturbing a previously normal scenario. But those disturbances are rare. As a result, it is reasonable to assume that the economy works coherently well in each state of nature as if they were worlds apart. When a sudden and unexpected particular event occurs—e.g., a financial shock—the economy may move from one state to the other with a positive probability. Switching models allow to estimate both the relationships between the variables in each state and also the probabilities of moving from one state to the other. Besides, stochastic shocks may vary over states. 2.5 Model Specification A four-variable MS-VAR (Markov switching vector autoregressive) model is implemented. In particular,21 (5) 𝑦𝑡=[F I r P ] where F is the financial conditions index (FCI)22 calculated in Section 2.1, I is the private investment23 growth rate, r is the nominal 1-day interbank interest rate (call rate) and P is CPI inflation excluding fruits and vegetables, Government-fixed prices and maid services (hereinafter, core inflation). All variables are quarterly, seasonally adjusted, standardized24 and expressed at annual rates. The data span from 1998Q3 to 2016Q2. 21 In earlier versions, private consumption and unemployment were used as proxies for real activity. By choosing private investment, the model was reduced to a four-variable VAR. 22 It would have been unwise to use the stress condition index (SCI) elaborated by BCU because of its short sample size (some series begin in December 2002). As a result, we calculate our own FCI from 1998Q3. 23 As a measure of activity; sensitivity analyses were performed using private consumption as a measure of wellbeing. It was decided not to use real GDP because it had been used to purge the variables included in F. 24 Standardization is crucial in order to assure a better measure of the relative loadings of each variable in the factors and to improve the reading of IRFs. 15 For identification purposes,25 I use the well-known Choleski decomposition, which imposes a recursive causal structure from the top variables to the bottom variables but not the other way around. The first variable is F because, by construction, the financial conditions index is independent of the business cycle variables and should just reflect pure financial shocks. The Uruguayan inflation data-generating process has proved to be a rather complex one and, at least in the short run, inflation does not seem to be exclusively a monetary phenomenon.26 Rather, inflation seems to be contemporaneously influenced by different factors stemming from financial, real and monetary markets. Then, P goes in the fourth place. Although the call rate has been the monetary policy instrument in Uruguay for only a short time (2007Q3-2013Q2) it is still the most reliable variable for monetary policy analysis. As such, monetary decision making does not seem to react contemporaneously to changes in private investment or current inflation rate;27 instead, the call rate seems to react contemporaneously to financial shocks and to real activity changes (through changes in private investment). Private investment is only affected contemporaneously by financial conditions and not by monetary policy decisions; a change in the 1-day interbank nominal interest rate does not reach the public immediately but with a lag, once interest rate arbitrages have been made and investment plans redefined. So, I goes in the second place and r goes in the third place. As a result, the chosen ordering is [F, I, r, P]. Following Hubrich and Tetlow (2014), there are some questions that are intended to be answered by this model. The first is whether whether there are periods of financial turbulence that appear randomly in the middle of normal times. Second, if such periods exist, which kind of regime switching better describes the data, that is, differences in long-run average growth rates (switching mean) and/or differences in the economic environment (switching variance). A third question is whether regime switching appears only in a specific equation—the stress equation alone or the monetary policy response to the financial stress—or in more than one equation of the VAR. 25 Impulse-response analysis must be done with the structural shocks; the ones recovered from the reduced form VAR are a weighted average of pure shocks. 26 See Bucacos and Licandro (2003). 27 It reacts to expected inflation deviations from the target. 16 3. Estimation This section presents the estimated model and discusses the results achieved, analyzing critically the performance of the estimated model in explaining the interelationship between some financial events and the Uruguayan economy. 3.1 Main Results The model presented is (6) 𝑦𝑡(𝑆𝑡)=𝜈(𝑆𝑡)+𝜋1𝑦𝑡−1+𝜀𝑡, 𝜀𝑡 ~𝐼𝐼𝑛�0, Σ(𝑆𝑡)� and (5) 𝑦𝑡=[𝐹 𝐼 𝑟 𝑃 ] It is estimated maximizing the multivariate log-likelihood function. The optimal lag length is 1. Different combinations of change of regime were estimated: only in the mean, only in the variance, only in the variance with switching scale, only in the variance with switching variance, and the combinations among them.28 Information criteria, mainly Schwartz criteria—and maximization of loglikelihood function are the goodness-of-fit selectors. The main results are displayed in Table 4, I and II. Table 4-I. Main results of MS-VAR Estimation Model Log L AIC SC N obs. N par. FAVAR(1) -375.56600 11.6342 12.2871 68 34 MSI(2)-VAR(1)1 -373.186927 12.0349 13.2099 68 36 MSIH(2)-VAR(1)2 -306.691667 10.1968 11.5024 68 40 MSIH(2)-VAR(1) 3 -285.571681 * 9.7510 * 11.2535 * 68 46 Notes: 1: Markov-switching in intercept term with fixed variance 2: Markov-switching in intercept term with heteroskedasticiy (switching scale in variance) 3: Markov-switching in intercept term with heteroskedasticity (switching variance) Log L = log likelihood value, AIC = Akaike Information Criteria, SC = Schwartz Criteria N Obs = number of observarions, N par = number of parameters, ( * ) indicates the chosen model. 28 MS models with changing parameters, MSA-VAR, are left for a future investigation. 17 The chosen model29 distinguishes two regimes: R_0 = “turbulence” and R_1 = “normal times,” both in the long-run mean and in the variance (switching variance). According to Krolzig’s (1997) notation, that model can be named MSIH(2)-VAR(1). Table 4-II. Main Results of MS-VAR Estimation ML estimates of the MSIH(2)-VAR(1) model, with SV (1998Q3 to 2016Q2) F I r P Intercepts R_0: Turbulence -0.94703 -0.66874 0.79574 0.03181 R_1: Normal times 0.21019 0.12339 -0.18264 0.03634 Long-run mean R_0: Turbulence R_1: Nomal times -1.52220 0.33785 -1.88306 0.34745 1.74438 -0.40037 0.39051 0.44606 Coefficients F-1 0.37785 -0.19216 0.01747 0.04884 I-1 0.00861 0.64487 -0.02504 0.01089 r-1 -0.13221 -0.40303 0.54383 0.23829 P-1 -0.07700 0.03340 -0.03065 0.91854 Standard errors R_0: Turbulence 1.59358 2.14411 1.31053 1.17388 0.96623 0.10091 3.06060 0.51492 R_1: Normal times Source: Author’s calculations. Table 5. Transition probabilities MSIH(2)-VAR(1) model, with SV (1998Q3 to 2016Q2) Regime 0,t Regime 1,t Regime 0,t+1 0.6107 0.0760 Regime 1,t+1 0.3893 0.9240 Source: Author’s estimates. 3.2 Discussion The time series are plotted together with their fit in Figure 5. Based on the MSIH model, the contribution of the Markov chain to the series is apparent for the inferred regimes clearly describe two distinctly different financial environments inside which economic activity has taken 29 According to the maximum log likelihood value and information criteria. 18 place in Uruguay in recent times. TRegime 0, called R_0, tracks turbulence in financial markets and dates recessions of the Uruguayan economy quite well, while Regime 1, called R_1, refers to normal times. Figure 5. Time Series and Their Fit Figure 6 diplays the regime probabilities of the two regimes in the MSIH model. According to Table 5, both regimes are persistent: in normal times, there is a 92 percent probability of stability next period, but if something unexpected disturbs that tranquility there is a 60 percent chance for that financial unrest to continue. Figure 6. Regime Probabilities in MSIH(2)-VAR(1) F 1-step prediction Fitted 0 20 40 60 0 5 F 1-step prediction Fitted d4_inv 1-step prediction Fitted 0 20 40 60 -5 0 5 d4_inv 1-step prediction Fitted CALL 1-step prediction Fitted 0 20 40 60 0 2 4 CALL 1-step prediction Fitted d4_p 1-step prediction Fitted 0 20 40 60 0 5 10 d4_p 1-step prediction Fitted 0 10 20 30 40 50 60 70 0.25 0.50 0.75 1.00 P[Regime 0] smoothed 0 10 20 30 40 50 60 70 0.25 0.50 0.75 1.00 P[Regime 1] smoothed 19 Table 6 diplays the regime classification based on smoothed regime probabilities. We can recongnize several events as instances of turbulence in financial markets (in R_0): the Brazilian financial crisis in 1999, Argentina’s devaluation in 2001 and its financial and political crisis afterwards, Uruguay’s triplet crisis in 2002-2003 (balance of payments, banking and fiscal crisis), the subprime crisis in 2008 and the declining commodity prices that led to the Russian financial crisis together with the Brazilian crisis in 2014. A description of those events is presented next. Table 6. Regime Classification Based on Smoothed Probabilities30 Regime Period Quarters Average probability R_0 = “Turbulence” 1999Q3-1999Q4 2 0.993 2001Q3-2002Q4 6 1.000 2003Q3-2003Q4 2 1.000 2008Q4 1 1.000 2014Q4 1 0.984 Total: 12 quarters (3 years, 17,65%) with average duration of 2.40 quarters (around a semester) R_1 = “Normal times” 2000Q1-2001Q2 6 0.980 2003Q1-2003Q2 2 0.999 2004Q1-2008Q3 19 1.000 2009Q1-2014Q3 23 0.999 2015Q1-2016Q2 6 0.999 Total: 56 quarters (14 years, 82.35%) with average duration of 11.2 quarters (almost 3 years) Source: Author’s estimates. In 1994 Brazil put into practice a stabilization plan named after its new currency, the real. Despite the Real Plan’s success in controlling inflation,31 the new currency was overvalued, which negatively affected Brazilian external accounts. To make matters worse, Brazil began to suffer from financial contagion from the Asian crisis in 1997 and the Russian crisis in 1998. In 30 Smoothed probabilites take into account all the sample information. 31 In 1994, the year the Real Plan was implemented, Brazil’s annual inflation rate exceeded 900 percent; by the end of 1998, price increases were negative. 20 order to stop investors from withdrawing their investments from Brazil, the authorities raised interest rates, but that measure increased the fiscal deficit to 8 percent of GDP. Finally, in January 1999 Brazil devalued its currency and abandoned its pegged exchange rate that was implemented together with other measures in 1994 in order to attack hyperinflation. By that time, the Brazilian economy was already in recession. After 1999, Argentinian exports were negatively affected by Brazilian real devaluation and a considerable international revalorization of the British pound, which led to a revaluation of the Argentinian peso against its main trade partner, Brazil (30 per cent of total commercial flows) and the dollar zone (23 per cent of commercial flows). Fernando de la Rúa took office as President in Argentina on December 10, 1999, when the recession was already underway;32 the economic stability achieved by the Convertibility Plan33 had turned into economic stagnation. The possible solution, abandonment of the fixed exchange rate with a voluntary devaluation of peso, was considered political suicide at the time, and “quasi-currencies” appeared in order to fill a liquidity gap.34 Under those circumstances, Argentina quickly lost investors’ confidence and capital outflows increased. In 2001, people fearing the worst began to withdraw large amounts of money from their bank accounts, exchanging pesos for American dollars and sending them abroad, which caused a bank run. Then, the Argentine government imposed capital controls and deposit freezes on Argentinean nationals’ time accounts, current accounts and saving accounts, a set of measures popularly known as the “Corralito.” Those measures turned out to be highly unpopular, resulting in the Cacerolazo,35 followed by a government-declared state of siege; still, protests turned violent and fatalities occurred. iolence and fatalities eventually resulted. the climate got worse and turned violent and violent and even people died in the protests. Finally, the government collapsed, and on December 20, 2001 the President had to leave the Pink House by helicopter. The Convertibility Law was abolished on January 6, 2002. As many analysts have pointed out,36 it is fair to say that the 2002 banking crisis in Uruguay might not have occurred had Argentina not collapsed first. But this exogenous 32 In 1999, Argentinian GDP fell by 4 per cent. 33 The Convertibility Law, Nº 23.928, was sanctioned on March 27, 1991 and established a fixed parity of 1 US dollar = 10,000 Australes (convertible Peso). 34 Such as Bono Patacón and Bono Lecop. 35 It refers to the beating of “cacerolas” (saucepans) by the people at predetermined hours as a sign of protest against the government. 36 De la Plaza and Sirtane (2005) and Paolillo (2004), among others. 21 contagion to the Uruguayan financial sector was magnified by inherent weaknesses of the Uruguayan economy and its banking sector. In effect, by the end of 2001, the Uruguayan economy was characterized by weaknesses in public banks,37 a high level of foreign indebtedness—both private and public38—and economic stagnation resulting from the appreciated real exchange rate of the Uruguayan peso in relation to its major trade partners39 (Figure 7). Figure 7. Uruguayan Real Exchange Rate The crisis in Uruguay40 began in December 2001 when the Argentine government imposed the “Corralito” and then that spark was quickly spread to the Uruguayan economy through the financial links between the two countries. Banco Galicia Uruguay and Banco Comercial, which combined represented approximately 20 percent of total deposits within the Uruguayan system, were both owned by Argentinian financial groups. At the begining, there were only instances of limited non-resident deposits runs related to those financial institutions, but as the crisis in Argentina was developing, deposits withdrawals gradually increased in 37 As of December of 2001, the level of non-performinge loans of the two main public banks, BROU and BHU, was greater than that of the rest of the system (39.1 versus 5.6); while after-tax return on equity was minus 4.5 for the public banks versus minus 0.9 for the private banks. In addition, BHU, the country’s almost only provider of mortgage lending, was vulnerable to external shocks owing to substantial currency and maturity mismatches in its balance sheets: 77 percent of deposits were US dollar-denominated and were at short maturities, while 94 percent of its loans were long-term peso-denominated. BHU had almost 10 percent of total deposits within the system. 38 Total Government debt was 58 percent of GDP in 2001, of which 83 percent was denominated in foreign currencies. 39 According to BCU figures, the Uruguayan real exchange rate index (base 2010=100) reached a value of 109.8 by December 2001. 40 See De la Plaza and Sirtane (2005) for a more detailed description of the Uruguayan banking crisis. 0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 160.0 2000.01 2001.01 2002.01 2003.01 vs Argentina vs Brazil Global 28 Figure 8. Impulse Response Functions 8.1 Only one regime (Linear FAVAR) 0 1 2 3 2 4 6 8 10 12 14 16 18 20 Shock 1 to var 1 -.6 -.4 -.2 .0 .2 .4 .6 2 4 6 8 10 12 14 16 18 20 Shock 1 to var 2 -.3 -.2 -.1 .0 .1 .2 246 8 10 12 14 16 18 20 Shock 1 to var 3 -0.8 -0.4 0.0 0.4 0.8 1.2 24 6 810 12 14 16 18 20 Shock 1 to var 4 Response to Cholesky One S.D. Innovations ± 2 S.E. 8.2 Markov switching regimes49 (a) Regime 0 (unrest) (b) Regime 1 (normal times) 49 I am grateful to P. Zagaglia (2011) for sharing his code and to S. Frache for his help in adaptating it to this investigation. 0246810 12 14 16 18 20 -0.5 0 0.5 1 1.5 shock1 to var 1 0246810 12 14 16 18 20 -1 -0.5 0 0.5 1 shock1 to var 2 0246810 12 14 16 18 20 -1 -0.5 0 0.5 1 shock1 to var 3 0246810 12 14 16 18 20 -3 -2 -1 0 1 2 3 shock1 to var 4 0246810 12 14 16 18 20 0 0.5 1 1.5 2 shock1 to var 1 0246810 12 14 16 18 20 -0.5 0 0.5 1 shock1 to var 2 0246810 12 14 16 18 20 0 0.05 0.1 shock1 to var 3 0246810 12 14 16 18 20 0 0.1 0.2 0.3 0.4 0.5 shock1 to var 4 29 3.3 Accountability So far, this model has given some answers. First, there are periods of financial turbulence that appear randomly in the middle of normal times. Second, the kind of regime switching that better describes the data is the one that combines differences in long-run average growth rates (switching mean) together with differences in the economic environment (switching variance); third, regime switching appears in more than one equation of the VAR, revealing different responses to the financial stress depending on the regime. 4. Conclusions This paper considers the effects of financial stress on the macroeconomy with the underlying idea that financial shocks affect the macroeconomy differently during “normal times” than during “stress.” The evidence supports the hypothesis that stress events are episodic in nature, their exact occurrence is unknown beforehand and they have a non-negligible probability of appareance. This behavior is captured in a multivariate framework through a Markov-switching vector autoregressive (MS-VAR) model. The comovements of a wide range of financial variables were summarized in one factor as a measure of financial instability. More precisely, F was calculated as an indicator of frictions, stresses and strains that occur in the financial system as a whole. There is evidence that a single-regime model is inadequate to describe the dynamics of the Uruguayan economy during the period of 1998Q3-2016Q2: there seem to be shifts in the stochastic shocks and the switching appears in all equations but those in price formation. Most of the sample corresponds to normal times, but around 17.65 percent of the sample corresponds to stress episodes, specifically in 2001Q3-2002Q4, 2003Q3-2003Q4, 2008Q4 and possibly in 1999Q3-Q4 and 2014Q4, with almost three quarters of average duration. During the “bad” regime (R_0), together with financial stress a negative pattern can be observed in the rest of the variables: a great swing in private investment and less stringent monetary policy.50 During the 50Also, according to sensitivity analysis done, a great swing in private consumption and higher unemployment is found during stress. Then, in normal times, private consumption stabilizes and unemployment falls. However, recovery from the crisis began before the actual crises ended. As a result, the long-run mean growth rate of consumption is higher during the “bad” regime than during the “good” one. In effect, efforts to overcome the 20012003 crisis began in the middle of it and by 2003Q2 the consumption growth rate had just offset the previous fall and all subsequent positive growth led to a positive average consumption growth rate for the 2001Q3-2003Q4 period. 30 “good” regime (R_1), however, the financial sector moves smoothly, private investment stabilizes and monetary policy tightens. Both situations (or states) seem to be quite persistent, for once the economy is there the probability to remain in the same state is high. For instance, if there was financial turbulence in the current period there is a 61 percent probability of financial stress in next period, and if the present period was a normal one, the probability of remaining in tranquility is 92 percent. During the bad regime, all shocks have larger variance than in the normal regime but the (pure) financial one. Besides, there is evidence of changes in the magnitude of the covariances among the variables.51 That is to say, during a stressful period, financial shocks have a negative mean and are more concentrated and are negatively linked to the other variables in the system. As a result, there are different patterns for the transmission of financial shocks. In effect, during financial stress, a pure financial shock reduces investment growth rate for four months while its effect is positive and almost unnoticeable during normal times. So far, the evidence supports the idea pointed out by Hubrich and Tetlow (2014) that financial unrest is necessary to reveal the linkages between the financial world and the macroeconomy. Financial stress seems to have some effects on the Uruguayan macroeconomy. Monetary policy is directly affected by financial conditions and becomes more contractionary; besides, the policy rate reacts more aggressively in the presence of financial stress than during normal times. Although stress events do not seem to have a significant impact on inflation rates, both private investment and private consumption (not reported here) react as expected. As a result, the evidence points out that macroeconomic stability seems to improve wellbeing. The results achieved so far are exciting and promising, but they should be taken with caution. In particular, we believe that this line of research may be improved by incorporating a wider group of financial variables into the data set—such as variables related to the banking industry—in order to design a better measure of financial instability. But trying to expand the time period does not seem to be an easy task. Next, the availability of lower frequency macroeconomic data (monthly to begin with) would allow a more realistic analysis of the dynamic relationships between the financial conditions and the macroeconomy. Then, the model could include parameters affected by different regimes which could allow for a changing 51 Sensitivity analysis show that the covariance between financial stress and consumption is almost inexistent (- 0.0042) during tranquility while it jumps to almost 500 times during financial stress (-2.0131). 31 response pattern depending on the state in order to investigate whether there are shifts in the dynamic propagation of shocks. And last but not least, the use of Bayesian methods would be recommended in order to mitigate the small sample limitations we have encountered. 32 References Bai, J., and P. Perron. 2003. “Computation and Analysis of Multiple Structural Change Models.” Journal of Applied Econometrics 18: 1-22. Barrán, J.P., and B. Nahum, B. 1967-1978. Historia Rural del Uruguay Moderno: 1851-1914. Volumes 1-7. Montevideo, Uruguay: Ediciones de la Banda Oriental. Bernanke, B.S., and A. Blinder. 1988. “Credit, Money and Aggregate Demand.” American Economic Review 78(2): 435-439. Bernanke, B.S., M. Gertler and S. Gilchrist. 1998. “The Financial Accelerator in a Quantitative Business Cycle Framework.” NBER Working Paper 6455. Cambridge, United States: National Bureau of Economic Research. Bucacos, E. 2015. “Impact of International Monetary Policy in Uruguay: A FAVAR Approach.” Documento de Trabajo 003 – 2015. Montevideo, Uruguay: Banco Central del Uruguay. Borraz, B., and D. Gianelli. 2010. “Un Análisis de Comportamiento a Nivel de Agente de la Encuesta de Expectativas de Inflación del BCU.” Documentos de Trabajo 001 – 2010. Montevideo, Uruguay: Banco Central del Uruguay. Bucacos, E., and G. Licandro. 2003. “La Demanda de Dinero en Uruguay: 1980:1-2002:4.” Revista de Economía, Segunda Época 10(2): 59-94. Calstrom, C., and T. Fuerst. 1997. “Agency Costs, Net Worth and Business Fluctuations: A Computable General Equilibrium Analysis.” American Economic Review 87(5): 893-910. Canova, F. 2007. Methods for Applied Macroeconomic Research. Princeton, United States: Princeton University Press. Capurro, A. et al. 2010. “Transmisión de la Política Monetaria a Nivel Agregado.” Montevideo, Uruguay: cinve. Mimeographed document. Castle, J., D. Hendry and J.A. Doornik. 2008. “Model Selection When There Are Multiple Breaks.” Economics Series Working Paper 407. Oxford, United Kingdom: University of Oxford, Department of Economics. Céspedes, L., R. Chang and A. Velasco. 2000. “Balance Sheets and Exchange Rate Policy.” NBER Working Paper 7840. Cambridge, United States: National Bureau of Economic Research. Chiesa, P., P. Garda and M.J. Zerbino. 2004. “Efectos Reales de la Política Monetaria en Uruguay: Una Aproximación del Estudio de los Canales de Tasa de Interés y del Crédito 33 Bancario.” Montevideo, Uruguay: Universidad de la República, Facultad de Ciencias Económicas y de Administración. Graduate thesis. De Brun, J., and G. Licandro. 2005. “To Hell and Back. Crisis Management in a Dollarized Economy: The Case of Uruguay.” Documento de Trabajo 004 - 2005. Montevideo, Uruguay: Banco Central del Uruguay. De la Plaza, L., and S. Sirtane. 2005. “An Analysis of the 2000 Uruguayan Banking Crisis.” Policy Research Working Paper 3780. Washington, DC, United States: World Bank Doornik, J.A. 2013. “Econometric Analysis Using Markov-Switching Models.” PCGive 14. London, United Kingdom: Timberlake Consultants Ltd. Ferreira, M. 2007. “Mecanismos de Transmisión de la Política Monetaria en Uruguay: Una Aproximación al Canal de Tasas de Interés y del Crédito.” Montevideo, Uruguay: Banco Central del Uruguay. Mimeographed document. Gertler, M. 1988. “Financial Structure and Aggregate Economic Activity: An Overview.” Journal of Money, Credit and Banking 20(3): 559-588. Hamilton, J.D. 1989. “A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle.” Econometrica 57: 357-384. Hatzius, J. et al. 2010. “Financial Conditions Indexes: A Fresh Look after the Financial Crises.” NBER Working Paper 16150. Cambridge, United States: National Bureau of Economic Research. Holló, D., M. Kremer and M. Lo Duca. 2012. “CISS: A Composite Indicator of Systemic Stress in the Financial System.” Working Paper 1426. Frankfurt, Germany: European Central Bank. Hubbard, G.N. 1998. “Capital-Market Imperfections and Investment.” NBER Working Paper 5996. Cambridge, United States: National Bureau of Economic Research. Hubrich, K., and R.J. Tetlow. 2014. “Financial Stress and Econonic Dynamics: The Transmission of Crises.” Working Paper 1728. Frankfurt, Germany: European Central Bank. Kashyap, A.K., and J.C. Stein. 1994. “Monetary Policy and Bank Lending.” NBER Studies in Business Cycles 29: 221-256. 34 Kashyap, A.K., J.C. Stein and D. Wilcox. 1993. “Monetary Policy and Credit Conditions: Evidence from the Composition of External Finance.” American Economic Review 83(1): 78-98. Kliesen, K.L., M.T. Owyang and E.K. Verman. 2012. “Disentangling Diverse Measures: A Survey of Financial Stress Indexes.” Federal Reserve Bank of Saint Louis Review (September-October): 369-398. Krolzig, H-M. 1997. Markov-Switching Vector Autoregressions: Modelling, Statistical Inference and Application to Business Cycle Analysis. Berlin, Germany: Springer. Landaberry, M.V. 2015. “Modelos e Indicadores de la Situación de Estabilidad Financiera: Metodología y Aplicación.” Documento de Trabajo 010 - 2015. Montevideo, Uruguay: Banco Central del Uruguay. Landaberry, M.V., and M. Tubio. 2015. “Estimación de Indice de Precios de Inmuebles en Uruguay.” Documento de Trabajo 011 - 2015. Montevideo, Uruguay: Banco Central del Uruguay. Licandro, G., and J. Licandro. 2001. “Anatomía y Patología de la Dolarización.” Documento de Trabajo 003 – 2001. Montevideo, Uruguay: Banco Central del Uruguay. Licandro, G., and M. Mello. 2012. “Canal de Hojas de Balance en Uruguay: Acelerador Financiero, Freno o Ambos?” Documento de Trabajo 015 - 2012. Montevideo, Uruguay: Banco Central del Uruguay. Masoller, A. 1998. “Shocks Regionales y el Comportamiento de la Economía Uruguaya entre 1974 y 1997.” Revista de Economía, Segunda Época 5(1): 141-214. Modigliani, F., and M. Miller. 1958. “The Cost of Capital, Corporation Finance and the Theory of Investment.” American Economic Review 48(3): 261–297. Paolillo, C. 2004. Con los Días Contados. Montevideo, Uruguay: Editorial Fin de Siglo. Pesaran, M., and Y. Shin. 1998. “Generalized Impulse Response Analysis in Linear Multivariate Models.” Economics Letters 58(1): 17-29. Ponce, J. 2012. “Precios de Fundamentos para las Viviendas en Uruguay.” Documento de Trabajo 017-2012. Montevideo, Uruguay: Banco Central del Uruguay. Ponce, J., and M. Tubio. 2013. “Precios de Inmuebles: Aproximaciones Metodológicas y Aplicación Empírica.” Documento de Trabajo 011-2013. Montevideo, Uruguay: Banco Central del Uruguay. 35 Sosa, S. 2010. “The Influence of ‘Big Brothers’: How Important Are Regional Factors for Uruguay?” IMF Working Paper 10/60. Washington, DC, United States: IMF. Terasvirta, T. 1994. “Specification, Estimation, and Evaluation of Smooth Transition Autoregressive Models.” Journal of the American Statistical Association 89: 208–218. Tong, H. 1990. Non-Linear Time Series: A Dynamical System Approach. Oxford, United Kingdom: Oxford University Press. Vaz, D.E. 1999. “Four Banking Crises: Their Causes and Consequences.” Revista de Economía, Segunda Época 6(1): 29-346. Zagaglia, P. 2011. “A Small Structural VAR Package for Impulse Response Analysis.” Available at: https://www.mathworks.com/matlabcentral/fileexchange/34358-a-small-structuralvar-package-for-impulse-response-analysis 36 Annex 1. Data Mnemonics Description Source log dif Seas i_a_agro_30 Nominal interest rate offered to agro firms up to 30 days in UY pesos Banco Central del Uruguay N N N i_a_com_30 Nominal interest rate offered to commercial firms up to 30 days in UY pesos Banco Central del Uruguay N N N i_a_ind_30 Nominal interest rate offered to industrial firms up to 30 days in UY pesos Banco Central del Uruguay N N N i_a_agro_30m Nominal interest rate offered to agro firms for more than 30 days in UY pesos Banco Central del Uruguay N N N i_a_com_30m Nominal interest rate offered to commercial firms for more than 30 days in UY pesos Banco Central del Uruguay N N N i_a_ind_30m Nominal interest rate offered to industrial firms for more than 30 days in UY pesos Banco Central del Uruguay N N N itlup 30-days Uruguayan Tbills Banco Central del Uruguay N N N 10T 10-year US Treasury Bills rate FRED N N N FFR Federal Funds rate FRED N N N n_a_agro Number of loans in UY pesos to agro firms Banco Central del Uruguay N N N n_a_com Number of loans in UY pesos to commercial firms Banco Central del Uruguay N N N n_a_ind Number of loans in UY pesos to industrial firms Banco Central del Uruguay N N N n_a_30m Number of loans (more than 30 days) in UY pesos to firms Banco Central del Uruguay N N N n_a_30 Number of loans (less than 30 days) in UY pesos to firms Banco Central del Uruguay N N N n_f_30m Number of loans (more than 30 days) in UY pesos to families Banco Central del Uruguay N N N n_f_30 Number of loans (less than 30 days) in UY pesos to families Banco Central del Uruguay N N N n_a_d_agro Number of loans in US dollars to agro firms Banco Central del Uruguay N N N n_a_d_com Number of loans in US dollars to commercial firms Banco Central del Uruguay N N N n_a_d_ind Number of loans in US dollars to industrial firms Banco Central del Uruguay N N N n_a_d_30 Number of loans (less than 30 days) in US dollars to firms Banco Central del Uruguay N N N n_a_d_30m Number of loans (more than 30 days) in US dollars to firms Banco Central del Uruguay N N N 37 Mnemonics Description Source log dif Seas n_f_d_30m Number of loans (more than 30 days) in US dollars to families Banco Central del Uruguay N N N n_f_d_cd Number of loans in US dollars to families on credit card usage Banco Central del Uruguay N N N mor_d_pdo Change in overdue accounts in US dollars, private sector with banking system Author´s own calculations on Banco Central del Uruguay data N Y N PWheat Commodity Price of wheat in US dollars; deflated by US CPI and deseasonalized International Monetary Fund Y Y Y PSoybean Commodity Price of soybean in US dollars; deflated by US CPI and deseasonalized International Monetary Fund Y Y Y PFood Commodity Price of food in US dollars; deflated by US CPI and deseasonalized The World Bank Y Y Y POil Commodity Price of oil in US dollars; deflated by US CPI and deseasonalized International Monetary Fund Y Y Y EMBI_URU Uruguayan country risk indicator República AFAP N N,Y N D_TCN Nominal depreciation Author`s own calculation on Banco Central del Uruguay data Y Y Y D_2_TCN Nominal volatility Author`s own calculation on Banco Central del Uruguay data Y Y Y D_TCR Real effective depreciation Author`s own calculation on Banco Central del Uruguay data Y Y Y VIX Chicago Board Options Exchange index volatility FRED res_to_gdp_ar Total reserves (excluding gold) to GDP ratio, for Argentina FRED N N N cr_to_gdp_ar Total credit (to non financial sector) to GDP ratio, for Argentina FRED N N N i 1-day interbank nominal interest rate Banco Central del Uruguay N N N C Private consumption Banco Central del Uruguay Y Y Y 𝜇 Unemployment rate Instituto Nacional de Estadística N Y Y P Core inflation Banco Central del Uruguay Y Y Y gdp Uruguayan Gross domestic product Banco Central del Uruguay Y Y Y P Uruguayan Gross domestic product deflactor Banco Central del Uruguay Y Y Y