The impact of exchange rate policy changes of the Bank of Russia on the economy of Armenia

Transcription

1 ՀՀ ԿԵՆՏՐՈՆԱԿԱՆ ԲԱՆԿ The impact of exchange rate policy changes of the Bank of Russia on the economy of Armenia Erik Vardanyan American University of Armenia February 2017 Abstract In this work, we examine the possible impact of the foreign central bank exchange rate policy changes on the economy of Armenia. First of all, we show that a significant increase of volatility of the value of ruble has taken place during the last decade. We believe that this was caused by changes in exchange rate policy conducted by the Central Bank of Russia. Afterwards, we estimate the possible consequences of these switches on the macroeconomic indicators of Armenia. By using VAR and ARCH types of models we reveal that the influence of the Russian economy on the Armenian economy is strong. However, the Armenian economy does not inherit the magnitudes of external economic variables movements while inheriting their directions. In other words, the increasing volatility of exchange rate of ruble does not enhance the volatility of the economic variables of Armenia. Key words: exchange rate policy, VAR, ARCH-GARCH models.

2 Acknowledgements I would like to gratefully acknowledge the faculty of Manoogian Simone College of Business and Economics at American University of Armenia, especially my thesis advisors Prof. Vardan Baghdasaryan and Prof. Gayane Barseghyan for their assistance in writing the master thesis. They were always available to discuss issues related to the research and to provide valuable comments. Also I would like to thank the chair of AUA M.S. in Economics program Prof. Vahram Ghushchyan for peer reviewing the research proposal, as well as for giving advices on how to write an academic style paper. All remaining errors are mine. 2

3 Contents 1.Introduction 4 2. Background 5 3.Data..6 4.Method 8 5.Estimation Results and Discussions 5.1 A hypothesis of increasing volatility of NEER of Russia VAR estimation Testing for possible impact of increasing volatility of exchange rate of ruble on the economy of Armenia Conclusion. 17 References Appendix

4 1. Introduction In today s world of fast globalization and deep economic integration of countries, unfavorable volatility in the domestic economy can be often caused by an outer trigger. Here emerging markets with underdeveloped financial systems are usually in more vulnerable position than advanced economies. The reason behind this could be the absence of enough efficient tools to overcome adverse financial shocks they face. Economic literature is rich with studies where the impact of real economic shocks on a particular economy is investigated. Often, they are about causes and consequences of supply shocks, cost-push or demand-side shocks. At the same time, relatively few studies are done about effects of abrupt changes in policies conducted by major state institutions like governments or central banks. In other words, often these kinds of actions themselves can be sources of instability not only in domestic economy but also in neighboring country s economies. In this sense, the possible impact of the actions of the CB of Russia on the Armenian economy is of particular interest for us. Namely, we investigate the impact of increased volatility of nominal effective exchange rate (NEER) of Russia on the Armenian economy. We believe that during time, the CB of Russia has significantly changed its exchange rate policy. This hypothesis is based not only on the visual examination of the data, but also, on the information International Monetary Fund (IMF) reports contain. In this study, we test whether the considered data reveal these changes. Also, considering the fact of strong economic connections between Russia and Armenia 1 we examine possible consequences of these policy shifts for the Armenian economy. The estimation section of our work consists of three parts. In the first one, we show that there were indeed significant switches in the exchange rate policy of the CB of Russia during the considered period. Our data investigation reveals that after each major financial crisis the Bank of Russia starts to tolerate comparably higher exchange rate volatilities. At the next step, by using a VAR model we emphasis the presence of strong economic connections of the two countries, which may bring the abovementioned increased uncertainty to the Armenian economy. Finally, in the third part of the work we 1 National Statistical Service of the Republic of Armenia discloses that about 25% of annual import of Armenia comes from Russia; IMF reports that 89% of remittances to Armenia are sent from Russia while on average the volume of remittances is about 16% of the value of Armenian national GDP 4

5 examine whether this kind of policy change spills over to the Armenian economy via the exchange rate channel. Analyzing results of the estimations, we come to the conclusion that though the Russian economy strongly affects the economy of Armenia, the latter does not inherit magnitude of the volatility of Russian financial variables. We consider this finding as our main contribution to the existing literature. The results of this research could be interesting to macroeconomists and particularly, to monetary policy specialists. 2. Background As we have already said, in the world of deep economic integration, abrupt monetary policy changes in large economies can create unfavorable fluctuations in economies of their trade partners. Several studies of similar economic developments are done. Among them, in (Mavromatis 2012) it is shown that there was a significant change in monetary policy reaction function of the Federal Reserve System of the USA around period of , which increased the volatility of monetary policy objective variables in Euro area while the European Central Bank did not alter its behavior. Although the originality of Mavromatis work inspires the actual research, the empirical results of our research show that the proposed econometric technique may not be applicable in the case of analyses of emerging markets. The author uses structural VAR model to capture contemporaneous relationships between macroeconomic variables of the advanced economies. Considering that the financial systems of these economies are well developed, most economic factors co-move and respond simultaneously to changes in economic conditions. On the other hand, in emerging markets, the speed of adjustment of economic factors is different. Here some variables respond contemporaneously to a shock, while for others, it may take a few more lags to adjust. Therefore, we find that while examining emerging markets, application of models of simultaneous estimation may often reveal misleading results. Furthermore, as it is shown in (Frankel 2010) and (Mishkin and Savastano 2000), there are substantial differences in transmission mechanisms of monetary policies in developed and emerging economies. This circumstance also should be considered in similar studies. Among others, high dollarization, high level of exchange rate pass-through, lack of credibility and underdeveloped financial systems are the main difficulties monetary policymakers of emerging markets are facing. Therefore, the 5

6 techniques we apply here may differ from those applied for similar studies of advanced economies. Here we define some concepts to avoid possible misunderstanding. By the monetary policy switch, we mean the change of the level of tolerance towards volatility of inflation, output gap or other objective for monetary policy variable. For example, a simple monetary policy rule could be represented as: (2.1) Where is the nominal interest rate,, are the monetary policy objective variables: inflation and output gap respectively, St is a random factor that evolves stochastically and independently of the endogenous economic variables, is exogenous policy disturbance; a and b are parameters set by policymakers, which define the stance of conducted monetary policy. These parameters indicate the level of the monetary policy tolerance; the higher they are, the less tolerable are for the monetary authorities the objective variables fluctuations. The objective variables could be functions of other economic factors. As it is shown in (Davig and Leeper 2008), the monetary policy switching framework implies that the coefficients of inflation and output gap in the reaction function could depend on exogenous factors (St), like lagged inflation or agents rational expectations. 3. Data For the actual research, we take monthly data from 2006 M1 to 2016 M8. The utilized variables are the monetary policy short-term interest rate (repo rate), Gross Domestic Product gap, consumer price index (CPI), nominal effective exchange rate (NEER) and international oil price. Therefore, we have four variables for both countries plus a variable representing oil price. The choice of variables is driven by the nature of our research and corresponds to similar studies (Dabla-Norris and Floerkemeier 2006) and (Bordon and Weber 2010). We select the repo rate, which is the key short-term interest rate used by CBs to signal their monetary policy stance. Taking into account the absence of monthly GDP data, we use industrial production index as a proxy to GDP (Mavromatis 6

7 2012). We use NEER instead of Real Effective Exchange Rate (REER) since it makes easier to distinguish exchange rate channel from other channels. An increase of NEER shows an appreciation of the domestic currency. Simply saying, NEER is an unadjusted weighted average rate at which country s domestic currency exchanges for a basket of multiple foreign currencies. Taking into account the significant dependence of Russian economy on prices of hydrocarbons (Sabitova and Shavaleyeva 2015), we include in our estimations a variable representing international oil price (Europe Brent Spot Price) as an explanatory variable. The repo rates are taken from databases of corresponding countries central banks. CPI, industrial production index and NEER of both countries are taken from the IMF database. Data on oil price were taken from the U.S. Energy Information Administration (EIA) database. The variable representing GDP gap in our research is a percentage deviation of actual GDP value from its potential. Since the potential is not visible, we use common Hodrick-Prescott (HP) filter technique to derive it from the actual data. Considering the time-series nature of our data and further estimations to be applied, the data should meet stationarity requirements. That is the reason for us to conduct a unit root Augmented Dickey-Fuller (ADF) test for each of the variables. The tests reveal that CPIs, NEERs, policy rates and oil price have unit roots and are difference stationary. Thus, we take first differences of these data. Before that, for explanatory easiness, we take natural logarithms of these data (except for policy rates). Consequently, in our research in case of small changes the variables representing NEERs and oil price are time-series of their percentage changes. Furthermore, the same transformation applied to CPIs make them approximately equal to inflation rates of corresponding economies. Therefore, henceforth instead of CPI we use the term Inflation rate for this variable. Additionally, since the economies of both countries are suffering from seasonality effects, some data are seasonally adjusted. Particularly, a seasonal adjustment Census X-12 technique is applied to GDPs and CPIs. The summary description of the data used can be found in the Table1 in the Appendix. 7

8 4. Method First of all, we empirically test the hypothesis whether the exchange rate policy of the CB of Russia has been changed during the considered period. For this purpose, we assume that the monetary policy of Russia is working under the inflation targeting framework and the CB commits to the Taylor-type monetary policy rule. The weakness of this assumption is the fact of the Russian economy being an emerging one, which is prone to supply and cost-push shocks (e.g. like fluctuations of international oil price). Therefore, in practice it is more plausible that the CB of Russia would conduct discretionary policy rather than may effectively commit to a certain rule. However, in our investigation this aspect will have minor influence on our outcomes. Since in considered policy rule function, we are mostly focused on dynamics of the relationship between policy rate and exchange rate. Here estimated parameters of output gap and inflation would have secondary importance. Furthermore, consideration of augmented Taylor-type monetary policy rule is a more formal way of estimation. In addition, as it is mentioned in (Frankel 2010), most emerging economies are suffering from high exchange rate pass-through. It is an economic phenomenon, which characterizes an impact of exchange rate changes on inflation rate. Simply, it shows how many percent of changes in domestic inflation is due to a percentage change in exchange rate. The higher is a country s dependence on import and international trade, the higher exchange rate pass-through could be expected. Consequently, a consideration of an exchange rate term as a Russian CB s monetary policy objective variable is justified. Our hypothesis of changes in the exchange rate policy of the Bank of Russia comes not only from visual examination of the data plotted (Graph 1), but also from IMF and the CB s reports. Thus, IMF Annual Reports on Exchange Arrangements and Exchange Restrictions for 2012 and 2014 (IMF Report, AREAER 2012), (IMF Report, AREAER 2014) reveal that during time the Bank of Russia gradually widened the band of allowable fluctuations of exchange rate and reduced amount of interventions. Additionally, in November 2014 the Bank of Russia officially announced about shifting to the floating exchange rate regime 2. Taking into account substantial dependence of the Russian economy on price of hydrocarbons (Sabitova and Shavaleyeva 2015), one may conclude that exchange rate 2 Details can be found on official web-site of the Bank of Russia: 8

9 fluctuations really matter for the Russian monetary policy. Thus, here we use augmented Taylor-type rule described in (Taylor 2001), which is often called an open-economy monetary policy rule. The rule has the following form: (4.1) Where is the policy rate, is the inflation rate, are the parameters determined by monetary policymakers, is the output gap, is the exchange rate term, is the error term. Our hypothesis is that the parameter is not stable and we expect that its value significantly changes during the considered period. To test this hypothesis, we apply the Wald test to the coefficient A3 of the estimated monetary policy rule function. Moreover, two specifications of the test are considered: with unknown dates of possible breaks and with known dates. In the first case, we assume that we do not have information about possible dates of structural breaks, and the test itself suggests most possible dates of breaks. On the other hand, the usage of the test with known date specifications allows us to test our suggested breaking points. These hypothetical dates were chosen based on exante information about stance of the Russian economy during the considered timespan. We believe that dates around major financial crises could be breaking points we are interested in. The outcomes of the tests with the alternative specifications reveal approximately the same dates of breaks. Therefore, based on these breaking points we separate the data into three subsamples and make corresponding estimations of the same policy rule function to find out the values of for different time-periods. In the next part of the analysis, we test whether our data depict economic connections of the two countries, claimed by previous empirical works. We need this to have a solid base to state that significant changes in the exchange rate policy of Russia can spill over to the Armenian economy. Here we use reduced-form VAR model to assess relationships between macroeconomic variables. We apply a reduced-form VAR approach proposed by (Sims 1980), as a regression of some vector of variables on lags of this vector. The matrix form of the model looks like: 9

10 (4.2) where a is a vector of constants, are matrices of coefficients and is a vector of innovations, which are serially uncorrelated disturbances with zero mean and variancecovariance matrix = Following the logic of similar studies (Dabla-Norris and Floerkemeier 2006) (Bordon and Weber 2010) we place the variables in VAR estimation by order from the most exogenous to the lest ones. First of all, taking into account the sample size and precision of the further estimation, it is appropriate to find out what variables from considered ones (except those representing the Armenian economy) could be removed from estimation to preserve degrees of freedom. For this purpose, we apply the Granger causality test. Additionally, this test gives us empirical evidence of the Armenian economy being affected by macroeconomic variables of the Russian economy and by fluctuations of international oil price. The test suggests including in the VAR model the oil price and NEER of Russia (the Table 3 in the Appendix). Therefore, the vector of endogenous variables and their order in the considered VAR model are the following: (4.3) where subscripts stand for considered country and t indicates time. We order GDP gap before inflation rate since we believe that output adjusts more sluggishly. Additionally, most probably exchange rate of Armenia is more volatile and is affected by other variables, but the impact in opposite direction could be weaker. Following the economic intuition, the external economic factors affect the Armenian economy but not vice versa 3. So far, we have revealed empirical models for examination of the presence of increased volatility of and the existence of economic connections between the two countries. Now we turn to the last part of our investigation: the examination whether the mentioned increased volatility is inherited by the economy of Armenia. For this 3 Although here we follow order requirements set by previous empirical works, change of order of variables in our research does not significantly alter final results. 10

11 purpose, we use a model from an ARCH-GARCH family, since our hypothesis is about a correlation of volatilities. Here we take the variance of Armenian dram as a dependent variable. We focus mostly on the explanation of the conditional variance of the exchange rate of AMD since we assume that it affects most of other macro-variables in the emerging economy. Here we assume that its heteroscedasticity could be explained not only by its own lagged ARCH and GARCH terms but also by similar terms of other exogenous variables. In the ordinary ARCH model with multiplicative heteroscedasticity, exogenous variables themselves are used in specification part of the model. However, for the purpose of our research, we slightly alter this approach by replacing these variables with their ARCH terms. Additionally, we include in the model a dummy variable to capture the possible effect of crises. Consequently, our suggested econometric model has the following form: = Const (4.4) + where are the parameters to be estimated, is the variance of the dependent variable, is the GARCH term, is the ARCH term of corresponding endogenous (y) or exogenous (x) variable, is a dummy variable (1 if it is period of the Global Financial crisis or the Financial crisis in Russia in 2014; 0 otherwise), is the error term. Here is the notion on how the ARCH terms of exogenous variables were calculated. Since all the considered variables are already stationary (with constant means), any deviation from their means refers to time-conditional variations of the corresponding variable. Therefore, we regress each variable on a constant. Afterwards, we predict error terms of such estimations. Finally, squares of these errors are represented in the final model as corresponding ARCH terms. 11

12 5. Estimation Results and Discussions 5.1 A hypothesis of increasing volatility of NEER of Russia As we were saying, first of all we test the hypothesis of instability of the coefficient of the exchange rate term in the assumed monetary policy reaction function of the CB of Russia. In the Graph 1, where percentage differences of NEER of Russia are plotted, one may notice that exchange rate movements are diverging. Here three distinct periods can be mentioned. The first one, with low volatility, starts from the beginning of the time period considered and lasts up to the period of the Global Financial crisis 2008/2009. The second period, with moderate volatility, occupies the period between the two crises: The Global Financial crisis and so-called Russian Financial crisis started in The third one, with high volatility, is the period afterwards. Graph 1: Percentage changes of NEER of Russia (with three distinguished periods of volatility) We believe that after each major crisis the CB widens the corridor of exchange rate fluctuations and eventually it shifts to the floating exchange rate regime (the CB of Russia officially announced about switch to the flouting exchange rate regime in November 2014). One of the main reasons behind this could be the fear of depletion of the CB s international reserves (Aizenman and Sun 2009). Furthermore, it could be described as a case of the crisis-driven exchange rate floating. We believe that the CB of Russia gradually increases the weight of the policy rate, as an exchange rate policy tool, rather than relies on the FX interventions. Therefore, we expect that application of the Wald test (both with known and unknown breaking dates) would reveal breaking dates approximately around the end of the years 2008 and While testing without date 12

13 specification for a single breaking point, we find out that the coefficient of NEER of Russia in estimated policy rule function significantly changes at the end of While testing hypothetical breaking points (for our data January 2009 and January 2015) we reject the null hypothesis of no breaks at 1% significance level. Therefore, we assume that at these points (which are associated with the major financial crises) the levels of tolerance of the monetary policy towards exchange rate fluctuations have been changed. Here we separate the data into three subsamples and estimate the policy reaction function using these subsamples. The outcomes of the estimations are represented in the Table 2 in the Appendix. An important finding is that the parameter of the exchange rate term changes from the sample 1 (the period 2006M1-2008M12) to the sample 2 (the period 2009M1-2014M12). The signs of these coefficients are negative and therefore are in line with the economic theory. It is worth mentioning that the magnitude of this parameter is increasing. We conclude that it can be a sign that the CB changes its exchange rate policy tool. In other words, the CB decreases the usage of international reserves (by conducting foreign exchange interventions) to affect exchange rate movements and increases utilization of the short-term policy rate for the same purpose. The intuition behind could be the fear of depletion of CB s international reserves. The corresponding parameter for sample 3 (the period 2015M1-2016M8) is not significant. We think that the reason behind this is a shortage of time-series span of the last subsample. Based on the results of the above mentioned two exercises (the tests for possible breaking dates and the estimation of the policy reaction function for different subsamples) we assume that there is enough evidence to claim that the monetary authorities have changed the way the exchange rate policy has been conducted during the considered period. The increasing volatility of exchange rate displays a movement towards the floating exchange rate regime. Although FX interventions and policy rate could be considered as two alternative tools of a CB to affect domestic exchange rate, in practice their applications are different. One would expect from a CB s FX interventions to have strong effect on exchange rate and inflation, but to have weaker direct effect on other macro variables. However, this is not the case while using the policy rate as a tool of exchange rate policy. Since short-term rate is a key rate in an economy and it influences economic activity in the country, high volatility of it 13

14 can create uncertainty in the economy. Therefore, usually under the inflation targeting framework central banks of emerging markets conduct so-called opportunistic monetary policy (Gali, Clarida and Gertler 1999). In other words, facing a temporary shock, a central bank may abstain from using the policy rate for a financial market interaction. 5.2 VAR estimation Although we expect that Russia, as one of the main trade partners of Armenia, does influence the latter s economy and several existing studies claim so, we find it proper to check whether our data show the presence of such a relationship. Primarily we have conducted Granger causality test to the estimated VAR with all the variables considered in the study (Table 3 in Appendix). The main aim of this exercise is to reduce the number of variables in the final VAR estimation, because of the concern of losing degree of freedom and reduction of precision of the estimation. Based on economic intuition and the outcome of the Granger test we decide to keep in the objective VAR all variables representing the Armenian economy, plus the variables representing oil price and the NEER of Russia. Consequently, by using reduced-form VAR modeling approach we estimate relationships between these six factors. The result of this estimation can be found in Table 4 in Appendix of the work. We use standard information criteria to determine the lag length of the VAR. For explanatory easiness, usually we interpret signs rather than magnitudes of estimates since for stationarity purposes, policy rates of both economies are represented in terms of their first differences and simply are percentage point differences. In Table 4 we see that in the equation of GDP gap of Armenia the first lag of the policy rate has negative coefficient, which is in line with the theory. A policy rate hike contractionary affects output. Positive correlation of policy rate and inflation in Armenia makes us conclude that the CB conducts countercyclical monetary policy. However, we cannot say anything about effectiveness of this policy. In other words, we cannot answer the question: does the CB short-term interest rate hike eventually lead to a decrease in inflation rate? This is one of the limitations of our research. This limitation comes from the nature of the considered data: our data are in monthly frequency. As we have already mentioned, in emerging economies monetary policy objective variables may adjust sluggishly. Therefore, it is challenging to make a conclusion about the future outcome of 14

15 current monetary policy actions. One can try to overcome this problem by using quarterly data, but since by the logic of our investigation we focus on exchange rate movements (which are more volatile), we have taken monthly data. We conclude that the exchange rate channel of transmission mechanism of the monetary policy is strong. One of the reasons behind this (besides the exchange rate passthrough) could be the exchange rate expectations of people, which determine people economic behavior and in turn affect NEER movements. For the case of the Armenian economy, the exchange rate expectations have strong influence because income of large share of the society considerably depends on the remittances from Russia. We see that indeed exchange rate matters, while examining inflation in an emerging economy like Armenia. Thus, in the equation of inflation rate, domestic currency appreciation in the first lag is associated with decline in current inflation rate, since purchasing power of the domestic currency increases. At the same time, we believe that positive signs of the coefficients of lagged inflation and output in the equation of depict a picture of stances of economic activity in the country. Increasing economic activity comes along with both upward inflationary pressure and exchange rate appreciation. The positive relationship between international oil price and Armenian output replicates the finding revealed in (Ayvazyan and Daba'n 2015). Although Armenia is an oil importer, an increase in oil prices has positive effect on its economy. This is associated with positive effect of increased oil prices on the economy of Russia. Which in turn enhances trade between both markets and increases amount of remittances sent to Armenia from its northern trade partner. Graph 2: Percentage changes of NEERs of Russia (solid) and Armenia (dashed) 15

16 One of the main findings of this VAR estimation is a positive co-movement of exchange rates of dram and ruble. This is logical, considering the abovementioned statements of strong economic links between the countries. From the outcomes, we assume that 1% increase in continuous compounded rate (an appreciation) of Russian ruble leads to 0.184% appreciation of Armenian dram. Based on the strong dependence of the stance of the Armenian economy on exchange rate, we believe that uncertainty and high volatility of exchange rate of domestic currency would eventually spill over to other macroeconomic factors of the economy. That is the reason why in the next section we focus on explanation of variation of the value of Armenian dram as a possible measure of uncertainty in the economy and try to examine possible impact of external factors on it. 5.3 Testing for possible impacts of increasing volatility of exchange rate of ruble on the economy of Armenia So far, by using historical data, we have proven the presence of increased volatility of exchange rate of the Russian currency during the considered period and have shown that NEER of Russia affects macroeconomic variables of the Armenian economy. Anyway, the question remains: does increasing volatility of the value of ruble enhance the volatility of Armenian economic variables? To test this hypothesis, we use an ARCH- GARCH family model with multiplicative heteroscedasticity specification. For this purpose, we include time-conditional variance terms of relevant factors in ARCH- GARCH model as explanatory variables. Also, it is worth mentioning that the Lagrange multiplier test reveals that indeed has heteroskedasticity issue. So here we try to explain the time-varying feature of its variance by exogenous factors. Taking into account sensitivity of ARCH-GARCH models to the choice of explanatory variables, we estimate the model with highest explanatory power (based on Akaike (AIC) and Schwarz (BIC) information criteria). The described econometric model has the following form: Var ( ) = Const. + + Var ( ) (5.1)

17 where Var () stands for variance term, ARCH () is the time-conditional ARCH term of the corresponding variable, is a dummy variable (1 if it is period of the Global Financial crisis or financial crisis in Russia in 2014; 0 otherwise), t is time operator, are the parameters to be estimated, is error term. The estimation of this model displays that the hypothesis of dependence of the variance of Armenian NEER on the variations in NEER of Russia is rejected (Table 5 in Appendix). The volatility of the exchange rate increases around the periods of huge financial turmoil (Global Financial crisis 2008/2009 and Russian Financial crisis in 2014), since substantial structural changes and increasing uncertainty take place during the periods of international crises. Thus, significant positive coefficient of the dummy variable is what we would expect. In Table 5, one may see that ARCH terms of policy rate of Armenia and lagged inflation rate have explanatory power in the estimated equation of variance of ; and their estimates have positive signs. One of the possible explanations of this phenomenon could be monetary policy actions of the CB of Armenia. We find that this is an evidence that in periods of high exchange rate and inflation volatility the CB actively uses its policy rate to stabilize the market. Gathering the results of VAR and ARCH-GARCH estimations we have concluded that in general our outcomes are in line with existing economic literature. However, the novelty of this research is the outcome that though the dependence of the Armenian economy on the Russian economy is strong, we do not have empirical evidence saying that the Armenian economy inherits magnitudes of movements while inheriting their directions. 6. Conclusion In the actual work, we have shown that the CB of Russia significantly has changed its level of tolerance towards exchange rate fluctuations twice during the considered period of time. Although the switches coincide with financial crises, changes of exchange rate movements are not temporary and they last far longer periods than the crises themselves. Evidences show that though the macroeconomic variables of Armenia are affected by the Russian economy, variations in NEER of Armenia are not explained by variations in the value of ruble. The Armenian economy inherits directions but not the magnitudes of Russian economic variables movements. One possible explanation for this 17

18 could be the presence of monetary policy actions of the CB of Armenia. The volatility of inflation and exchange rate are associated with the volatility of short-term policy rate. We find that the latter is used by the CB as a market stabilization tool in periods of increasing uncertainty. Speaking of limitations, among others we find that the research would be more profound if more relevant economic factors were included in the estimations. However, considering the short-run time series problem of emerging markets the task is not feasible. 18

19 References Aizenman, Joshua, and Yi Sun "The financial crisis and sizable international reserves depletion: From 'fear of floating' to the 'fear of losing international reserves'?" NBER Working Paper. Ayvazyan, Knarik, and Tereza Daba'n "Spillovers from Global and Regional Shocks to Armenia." IMF Working Paper. Bordon, Anna Rose, and Anke Weber "The Transmission Mechanism in Armenia: New Evidence from a Regime Switching VAR Analysis." IMF Working Paper. Dabla-Norris, Era, and Holger Floerkemeier "Transmission Mechanisms of Monetary Policy in Armenia: Evidence from VAR analysis." IMF working Paper. Davig, Troy, and Eric M. Leeper "Endogenous Monetary Policy Regime Change." NBER Working Paper series (University of Chicago Press) Volume ISBN: Frankel, Jeffrey A "Monetary policy in emerging markets: a survey." NBER Working Paper series. Gali, Jordi, Richard Clarida, and Mark Gertler "The Science of Monetary Policy: A New Keynesian Perspective." Journal of Economic Literature. IMF Report "Annual Report on Exchange Arrangements and Exchange Restrictions." Washington, D.C., 20. IMF Report "Annual Report on Exchange Arrangements and Exchange Restrictions." Washington, D.C., Mavromatis, Konstantinos "Markov Switching Monetary Policy in two_country DSGE Model." Warwick Economic Research. Mishkin, Frederic S., and Miguel A. Savastano "Monetary policy strategies for Latin America." NBER Working Paper series. Sabitova, Nadia, and Chulpan Shavaleyeva "Oil and Gas Revenues of the Russian Federation: Trends and Prospects." Department of Finance, Institute of Management, Economics and Finance, Federal University, Kazan, Russia (Procedia Economics and Finance). Sims, Christofer A "Macroeconomics and Reality." Econometrica 48 (no.1(january)): Taylor, John B "The Role of the Exchange Rate in Monetary-Policy Rules." American Economic Review

20 Appendix Table 1: Summary description of the data Vari ables Obs. Description Unit root 124 Percentage point difference Season. adjust ment Mean Std. Dev Min. Max. I (0) Percentage I (0) change 128 Percentage I (0) deviation from its potential 127 Percentage change I (0) Percentage point difference I (0) Percentage change I (0) Percentage I (0) e deviation from its potential 127 Percentage I (0) change Oil price 127 Percentage change I (0) Notes to the Table 1. Here industrial production index is used as a proxy of monthly GDP, and further it is used to calculate GDP Gaps of the economies. GDP Gaps are percentage deviations of actual GDPs from their potential levels. The potentials were calculated by using Hodrick-Prescott filtering technique (with parameter λ=14400). An application of a unit root test reveals I (1) variables: policy rates, NEERs and oil price. These variables in this research are represented in terms of their first differences. Therefore, all variables in this table are stationary. Furthermore, time series of the variables NEERs and oil price could be interpreted as corresponding percentage changes. Data on policy rates (repo rates) were taken from databases of the corresponding country s central bank. Oil price was taken from U.S. Energy Information Administration. All the other variables were taken from database of IMF. Inflation rates and GDPs are seasonally adjusted. Although in some similar studies exchange rates also are in seasonally adjusted terms, we do not transform the raw data, since our data did not display vivid seasonality patterns. In the table: + stands for application of seasonal adjustment to the data, while - otherwise. 20

21 Table 2: Estimation of assumed policy reaction function of the CB of Russia for three different periods (the dependent variable is policy rate of the CB of Russia) Variables 1 period Coefficients (2006m1-2008m12) 8.23 (5.71) (1.22) *** (4.55) 2 period Coefficients (2009m1-2014m12) (12.94) 9.72*** (2.99) *** (2.74) 3 period Coefficients (2015m1-2016m8) (14.18) 25.82** (10.54) (2.12) Note to the table 2. Here augmented Taylor-type policy reaction functions for Russia are estimated. The same model was applied for three different periods of time: 1 period starts from beginning of the data considered and lasts up to the Global Financial crisis; 2 period period between two crises: Global Financial and Russian Financial crises (starts in 2014); and 3 period starts from Russian Financial crisis and lasts up to nowadays. Breaking points of the data were found based on the results of the Wald test, which shows dates of possible structural breaks in the coefficient of NEER of Russia in the augmented Taylor-type reaction function (estimated using whole sample: 2006m1-2016m8). Correspondingly, 2009m1 and 2015m1 breaking points were found. The coefficients of the NEER are negative, which is in line with the theory. Their magnitudes increase during the period (except for the last period, where the outcome is not significant. We associate it with scarcity of observations for that period). Based on this outcome we conclude that the CB starts to rely more on the short-term policy rate as an exchange rate policy instrument. Standard deviations are reported in parentheses; stars stand for significance levels: ***- at 1%, **- at 5%, * - at 10%. 21

22 Table 3: Results of Granger causality test Dependent variable Exogenous variables Oil price Oil price Oil price Oil price p-value * ** *** 0.051* ** *** 0.059* Note to the Table 3. Here results of Granger causality test are represented, where Armenian economy variables are dependent variables. The null hypothesis of the test is no Granger causality. Namely, here is the technique of the test: it is a Wald test that coefficients of two lags of particular independent variable in estimated VAR are jointly equal zero. For example, in first row of estimated VAR, Armenian GDP gap among others are regressed on first and second lags of GDP gap of Russia. And the test does not reject the null of no causality. We see that output, policy rate and NEER of Armenia are significantly Granger-caused by NEER of Russia and by oil price. At the same time, despite the test suggests that Russian inflation rate affects Armenian exchange rate, we do not include Russian inflation rate in the further models to estimate. Since we believe that this connection is not direct and it goes via the exchange rate channel. Therefore, in further estimation, alongside with the Armenian variables we include NEER of Russia and international oil price. Stars stand for significance levels: ***- at 1%, **- at 5%, * - at 10%. 22

23 Table 4: Estimated reduced-form VAR (columns represent endogenous variables, rows are lags of variables) Variables Oil price L. Oil price 0.283*** (0.100) 0.070** (0.035) 0.122** (0.057) (0.008) (0.323) (0.029) L2. Oil price (0.096) 0.057* (0.034) (0.055) 0.022*** (0.007) 0.588* (0.312) ** (0.028) L (0.272) 0.563*** (0.095) (0.156) (0.021) *** (0.881) (0.080) L *** (0.270) *** (0.094) (0.155) (0.021) * (0.876) 0.184** (0.080) L (0.157) (0.055) 0.785*** (0.090) (0.012) (0.510) 0.099** (0.046) L (0.176) (0.061) 0.198** (0.101) (0.014) (0.571) (0.052) L (1.172) (0.408) (0.673) 0.286*** (0.091) (3.800) 0.873** (0.346) L (1.174) (0.409) (0.674) * (0.091) 9.729** (3.808) (0.347) L *** (0.028) (0.010) ** (0.016) (0.002) 0.278*** (0.090) (0.008) L ** (0.027) 0.017* (0.009) 0.031** (0.015) 0.004** (0.002) 0.145* (0.087) (0.008) L (0.317) (0.110) (0.182) ** (0.025) (1.027) 0.158* (0.094) L (0.312) (0.109) (0.179) 0.063*** (0.024) 3.529*** (1.011) (0.092) Constant (0.010) (0.003) (0.005) 0.003*** (0.001) * (0.031) (0.003) Number of 120 observations AIC -2, BIC -2, Note to the Table 4. In this table estimation of reduced-form VAR is represented. Here we include all four variables of the Armenian economy, plus international oil price and NEER of Russia (referring to the results of Granger causality test). Standard deviations are reported in parentheses; stars stand for significance levels: ***- at 1%, **- at 5%, * - at 10%. AIC and BIC stand for Akaike and Schwarz information criteria. L. is a lag operator. 23

24 Table 5: Variance estimation of NEER of Armenia ( is the dependent variable) Variables Coefficients Mean ( ) (0.001) HET (77.553) ** (0.731) *** ( ) 0.894* (0.486) Constant *** (0.221) (0.093) (0.082) Note to the Table 5. Here the variance of the variable representing exchange rate of Armenian currency is estimated. The ARCH-GARCH model with multiplicative heteroscedasticity specification is used. In the section named HET ARCH terms of different variables are used to explain heteroscedasticity of the dependent variable. The variable Crisis is dummy and has value 1 for periods of the Global Financial crisis and a financial crisis in Russia in 2014; 0 otherwise. As one may see the financial variable of Russia does not have explanatory power in the model. As it could be expected, in periods of crises variance of the dependent variable increases, but we cannot say anything about causality of the events, since during a crisis economic relationships and mechanisms may be different comparing to the crisis-free periods. Finally, we see positive co-movement of conditional variances of Armenian exchange rate, policy rate of Armenia and lagged inflation rate. We consider this as an evidence of the CB using its policy rate to stabilize the market in periods of high volatility of the inflation rate and the exchange rate (considering high exchange rate passthrough). Additionally, since ARCH and GARCH processes of the estimation do not have explanatory power, we conclude that the choice of exogenous variables is correct; these variables explain significant part of the conditional variance of the dependent variable. Standard deviations are reported in parentheses; stars stand for significance levels: ***- at 1%, **- at 5%, * - at 10% 24