Panel Data Analysis of Exchange Rate for Fragile Five

After the Bretteon Woods System, as a result of the preffering especially the flexible exchange rate systems of developed countries the exchange rate risk has emerged. The transfer for the flexible exchange systems in the majority of developing countries started with the financial liberalization in the 1990’s. In this period, the progress of information technology and globalization have rendered exchange rate policies and exchange rates a priority for countries and exchange rates have become effective on macroeconomic indicators. In this study was examined exchange rate behaviour of Morgan Stanley’s “Fragile Five Countries” which are Brazil, India, Indonesia, South Africa and Turkey. For this purpose with the Panel Cointegration Tests was investigated if there is long termed relationship between the exchange rate and international reserves, money supply and The Bloomberg U.S. Financial Conditions Index(BFCIUS). As well as, Granger Causality test is applied using Panel Data Causality Techniques are used to uncover the direction of relation between variables. Thus Panel Vector Autoregression (PVAR) Model was estimated among the variables. The present study is distinguished from previous studies by investigation of long term relationship between also The BFCIUS, exchange rate. Data base is containing from the exchange rate index, international reserve index, Money supply index and BFCIUS variables presented by Bloomberg and monthly data includes 1015 observations done in the period of March 2000 – January 2017.


Introduction
Exchange rate refers to the ratio of the currency of a country to the currencies of other countries.There is an interaction between the countries' exchange rates and economic activities.This interaction among the exchange rate policies and regimes arise from the need for fulfilment of the objectives such as price stability and financial stability and sustained competitiveness of a country.After being used for exchanging goods and services for hundreds of years, at the end of 19 th century the use of precious metals gave way to the era of "gold standard" in which currencies were pegged to gold.The use of "gold standard" prevailed until the First World War.In the period between the two World Wars, different payment systems were implemented although none of them proved successfully.
In 1944, towards the end of the Second World War, Bretton Woods Agreement was signed.In Bretton Woods System, also known as the adjustable peg system, when the price of USD was pegged to gold, member countries also pegged their currencies to USD over a fixed parity (Danış, 2008:25-35).As a result of increasing speculative attacts claiming that the value of USD could not be preserved, in addition to the growing deficits in USA's balance of payments, liberalization of private capital flows and the incurred oil shock(1973) etc., Bretton Woods System was abolished in 1973 (Pazarlıoğlu and Güloğlu, 2007:20).
As the nations failed to reestablish a new monetary system following the collapse of Bretton Woods, most of the developed countries preferred to use the floating exchange rate regime.In this period, as the volatility of exchange rates increased in international financial markets, new theories were developed to elucidate the nature of exchange rates.The first theory to explain foreign exchange rate flows is Purchasing Power Parity (PPP) Approach.Afterward Monetarist Approach was developed by economists such as R.Mundell, H.Johnson and J.Frenkel, R. Dornbusch in 1970s when monetary policies increased in importance.According to the Monetarist Approach, which is based on Mundell-Fleming approach, there is a complete flow of capital among countries Monetary models are supported with approaches that incorporate variables such as interest rate, income level, money supply and relative prices, into their analyses (Dülger 2002:49).Monetary models are built in consideration of both domestic and abroad supply and demand.These models can be categorized as flexible priced and fixed priced monetary models (Özkan, 2012:30).
In flexible priced monetary models PPP is always assumed to apply (Özkan, 2012:30).But according to fixed priced monetary models, PPP assumption only applies in the long term, and in the short term prices are always fixed.Frenkel (1976) Interest Rate Differential Model and Dornbusch(1976) over shooting effect theory are among the fixed priced monetary models.Monetary models are criticized on the grounds that they do not suffice in estimating foreing exchange rate fluctuations and it is stated that monetary instability is only one of the factors that dominate the volatilities in exchange rate (Calderon and Kubota, 2009:2).
Increasing globalization, capital mobility enabled simultaneous trading of foreign currencies in financial markets, while providing integration among financial markets by means of developing information technologies.In most of the developing countries, adoption of flexible exchange rate systems started with financial liberalization in 1990s.With increasing financial liberalization, exchange rates and exchange rate policies became more important for countries resulting in an increased interaction between exchange rates and macroeconomic indicators.Several alternatives were proposed on choosing between flexible and fixed exchange rate systems or building mixed models that incorporate the advantageous aspects of these two systems.
Currently several countries have been using the floating exchange rate system, including Turkey.In floating exchange rate regimes, exchange rates are defined on supply and demand basis.Adopted monetary and fiscal policies, economic foundations, projections and international conjuncture together define the supply-demand conditions for foreign currencies.On the other hand, intervention by central bank is allowed in case of extreme volatilities, imposed to restrict the risks against financial stability (TCMB, 2016:2).To a certain extent, foreign exchange reserves of central banks are also important indicators depending on the exchange rate regime.Cental Banks of developing countries, particularly those with fragile economies, hold serious amounts of foreign exchange reserves despite their high cost.This way, strengthening the country's economy against national and international crises, discharging public debts and fulfilment of other foreign currency requirements, building a higher level of confidence towards the country and sustainment of the monetaryexchange rate policies are aimed (TCMB, 2011:2-3).
Beside the fluctuation levels of exchange rates their interaction with macro-economic variables is also regarded as an important factors while building exchange rate systems.Accordingly, the relationship between the exchange rate volatility and macro-economic variables have been the subject of sevaral studies.Balg and Metcalf(2010) proposed that, exchange rate volatility is in essence highly affected by the volatility in money supply.Caporale et al. (2011), concluded that, the financial crises in South American countries arose from the changes in the real exchange rate.Güloğlu and Akman (2007) stated that, the exchange rate (TL/USD) volatility in Turkey was influenced by the political and economical developments in financial markets, which resulted in a permanent volatility.
Another field of study is the analysis of the effects of global financial crisis on the exchange rate policies of developing countries.In their study, Coudert et al. (2011) investigated the effects of global financial crisis on the exchange rate policies of developing countries.In this study, the volatility in exchange rates in 21 developing countries and global markets in the period of 1994-2009 were analyzed.According to the researchers' findings, exchange rate volatilities in the sample developing countries increased at a higher rate than the global financial stress.Such difference in the rate of increase was mainly ascribed to the weakened link between local currencies and USD, as a result of increased market pressure in countries following the crisis.Also, the volatility in the currency of one country in the sample was found to spread to other countries as a result of regional contagion.2011) reported that, foreign interventions of FED influenced the USD-English Pound, and USD-Swiss Frank rates.It was also stated that, intervention prospects were likely to cause increased transaction rates and volatility in exchange rates.As reported by Tunay (2008), Central Bank Republic of Turkey's (CBRT) intervention to currency markets with ARFIMA-GARCH and ARFIMA-FIGARCH methods between 1999-2008 increased the volatility in USD and EURO exchange rates.
The effects of FED's tapering talk on developing countries were investigated in some of the related studies.In their study, Basu et al. (2014) investigated the effects of FED's tigtening policy statements via tapering talk on Indian economy.In this period, five developing countries, also referred to as Fragile Five, reacted to FED's statements with utmost concern.In this period Indian Rupee underwent 18% depreciation.Indian economy's higher fragility among other countries was mainly attributed to its vulnerability against capital outflows as a result of its weak financial conditions.Against such financial crises, authors have proposed a number of solutions such as keeping higher levels of foreign currencies and making medium term plans so as to maintain the sustainability of current deficit.According to Eichengreen and Gupta (2014), FED's tightening statements via tapering talk in April-August 2013 period was effective on the exchange rates, foreign currency reserves and stock prices with direct proportion to the financial market size and liquidity in developing countries, leading to a serious level of reserve currency loss and depression in exchange rates.
Numerous studies have been conducted on the relationship between exchange rate and international curreny reserves.Ahmad and Pentecost (2009), applied the threshold cointegration technique developed by Balke and Fomby (1997) to investigate the relationship between a number of African countries' exchange rates and international currency reserves within 1980-2004 period.As a result, threshold co-integration was found between the series.It has been observed that there is a long dynamic relationship between the variables.Floating exchange rate regimes exhibit higher levels of threshold co-integration as compared to fixed exchange rate regimes.Evidently, exchange rates are adjusted more frequently than international currency reserves.Also, existence of a linear cointegration between the variables was refuted in the conducted study.
Determination of the strong interaction of international reserves with exchange rates has raised the importance of reserves.Because of this various criteria have been developed in the literature to determine the sufficiency of held reserves.Traditional ones among these criteria are categorized as the reserves / imported goods and services, reserves / short-term foreign debts, and reserves / total money supply.Although there is no consensus on the subject, the reserves / money supply ratio is generally expected to be at 20% level.This ratio also reveals the role of international reserves under the responsibility of banking sector (TCMB, 2011:4).

Memiş et al. (2014) applied
Johansen's co-integration test and Granger's casuality test for the analysis of the relatinship among CBRT's reserves and GDP, consumer prices index, foreign debt, real exchange rate using the quarterly data of 1989-2013 period.Briefly, foreign debt affects the exchange rate, and in turn, exchange rate affects amount of reserves.In this research, the authors propose suggest that the relationship between the exchange rate and the foreign debt should be taken into consideration by CBRT while establishing the monetary policy (Memiş et al., 2014:93-105) Çakmur Yıldırtan(2009) estimated long run relation between exchange rate and macro economic variables (interest rate, inflation, money supply, crude oil prices) in Turkey for the period 1994-2008.The result of cointegration test suggest there subsist a long run relation between exchange rate and interest rate also exchange rate and money supply.The results of VEC Granger Causality/Block Exogeneity Wald Test show that there is bilateral causality between Exchange rate and money supply also unidirectional causality from money supply to exchange rate and from interest rate to money supply.
The economies of Brazil, Indonesia, South Africa, India and Turkey show similarities in terms of indicators such as high inflation, high indebtedness in foreign currency, low growth levels and volatile capital flows.This group, which has similar structural problems, are affected by international market flows at higher rates than other counries."Fragile Five" countries term for this group of five countries initially used by Morgan Stanley (2013).National currencies of Fragile Five countries have undergone serious depreciation in recent years(Şahin, 2016:321).
A number of studies were conducted on the effects of foreign exchange rates on macro-economic variables in Fragile Five countries.Güvercin (2016) detected significant and assymetric effects of the changes in USD foreign exchange rates on stock market indices in Fragile Five countries within 2002-2015 period.Şahin (2016) investigated the sensitivity of share indices to USD exchange rates in Fragile Five countries using monthly data of the counties over a span of 20 years, and found that the variables in general exhibited a non-linear behavior with a partial unit root, using Caner and Hansen's test (2001).USD, EURO and GBP are the leading currencies used in the international financial markets, and values of these currencies with respect to others are determined within the frame of free floating exchange rate system (Güvercin, 2016:366).On the other hand, prevalence of USD among other currencies in currency based distribution of world foreign currency reserves prompted us to examine the value of USD against Fragile Five country currencies.
Determination of a correlation between exchange rates and financial variables and this correlation's level can be a guiding factor in estimation of the exchange rate risks.The dominance of USD in international trade and the fact that it constitutes a major part of the reserves hold in central banks(as seen from Figure 1), increases the sensitivity of country economies to the developments in USA economy.International reserve sufficiency can also be regarded as an important indicator of fragility.In the present study, the "international reserves" in Fragile Five countries, used to determine the reserve sufficiency in Fragile Five countries, is also studied.In recent years, the Financial Indicators (FCI) that developed by many organizations have been used.The common point of FCIs is to summarize key indicators of financial market conditions into a single number.There are two reasons why we prefer the Bloomberg US Financial Condition Index (BFCIUS) in our study.First, as seen in Table 1 it provides a measure of the daily weighted statistics of the money markets, equity markets and bond market in the United States.The second reason is that the general conditions of US sum of financial markets indicators are accepted as a true indicator.Therefore, Bloomberg U.S. Financial Conditions Index(BFCIUS).This research is distinguished among other literature studies with investigation of the long term relationship between BFCIUS and the exchange rate.In the present study, exchange rate behaviors of "Fragile Five" countries are investigated for the period between March 2000 and January 2017.It was examined as to whether there is a long term relationship among rates, international reserves, money supply and BFCIUS, using Panel Cointegration Tests.Also, Granger's casuality test and panel data

Methodology and Data
In this study in order to investigate the long run relations of exchange rate and international reserves, money supply and BFCIUS of the Fragile Five countries.Data collected from Bloomberg and Central Bank Republic Of Turkey(tcmb.gov.tr) and also E-views 8 was used to estimate the models.The study brings the cross-sectional panel data of Fragile Five countries, over the period of 2000:03 and 2017:01 with 1015 observations.
The cointegration analysis of panel data consists of four steps: First, we test for a panel unit root.

Panel Unit Root Test
According to the equation with the individual intercept and without intercept and trend of panel unit root tests of Levin, Lin and Chu (LLC), Im, Pesaran and Shin (IPS), Augmented Dickey Fuller Fisher and Philip Perron; the null hypothesis claiming 'blmindex' variable include 1% panel unit root is rejected.According to the test equation we found that; 'exrate' 'intreserves' 'moneysupply' are not stationary in their level values but became stationary in their first differences but 'blmindex' is stable in its own level value.'dexrate' 'dintreserves' 'dmoneysupply' variables symbolizes first difference of the series.The results are shown in Table 2.

Table 2. Panel Unit Root Test Findings
Values in parenthesis indicates the probability values relating to the test statistic.According to the significance level; for 5%; (*) for; 1% (**) According to the test statistic indicates that the relevant variables stationary.Lag length selection based on Scwarz Info Criterion with a max lag of 14.
Using these results, we proceed to test 'exrate' 'intreserves' 'moneysupply' for cointegration in order to determine if there is a long-run relationship to control for in the econometric specification.Becaueso of precondition for running Panel Cointegration Test variables should be non-stationary at the level values but they should be stationary in their diference.Because of the 'blmindex' is stable in its own level value we excluded from the Panel Cointegration Test.We have fulfill the condition of Panel Cointegration Test.

The Panel Cointegration Test
The panel cointegration tests Pedroni (1999) tests allow for heterogeneity among individual members of the panel, including heterogeneity in both the long-run cointegrating vectors and in the dynamics, since there is no reason to accept that all parameters are the same across countries.Two types of tests are recommended by Pedroni.The first type is based on the withindimension approach, which includes ; panel m-statistic, panel qstatistic, panel PP-statistic, and panel ADF-statistic.Those statistics pool the autoregressive coefficients across different members for the unit root tests on the estimated residuals.The other test by Pedroni is depend on the between-dimension approach, which includes; group q-statistic, group PP-statistic, and group ADF-statistic.Those statistics are depend on estimators that simply average the individually estimated coefficients for each member (Chiang Lee.2005).Once the three variables are cointegrated, next step is to implement the Granger Causality Test.We use a Dumitrescu Hurlin Panel Causality Test model to account for the long-run relationship.

Dumitrescu Hurlin Panel Causality Test
According to Granger (1969), the Granger causality means that the knowledge of past values of one variable (X) helps to improve the forecasts of another variable (Y).If there is cross-sectional dependency and heterogeneity across countries, the method utilized should account for these features.Since Granger Causality is computed by running bivariate regressions, there are a number of different approaches to testing for Granger Causality in a panel context.
Where "t" denotes the time period dimension of the panel, and "i" denotes the cross-sectional dimension.
Determining whether slope coefficients are homogeneous or heterogeneous is also important in a panel causality analysis to impose causality restrictions on estimated coefficient Dumitrescu-Hurlin (2012), makes an assumption, allowing all coefficients to be different across cross-sections: 'Dumitrescu Hurlin Panel Causality Tests' for evaluating the causality relationships between the variables.This test is calculated on the basis of conventional Granger causality regressions for each cross-section individually.This test is calculated by simply running standard Granger Causality regressions for each cross-section individually.The nest step is to take the average of the test statistics, which are termed the Wbar statistic.They show that the standardized version of this statistic, appropriately weighted in unbalanced panels, follows a standard normal distribution.This is termed the Zbar statistic.The significance of 'W statistics' and 'Zbar statistics' show the rejection for null hypothesis of 'no causality', and show the following three possibilities; if two variables have a unidirectional causality between the variable or if there is a bidirectional causality running between them, and if there occur no causality.This test can be used when N is growing and T is constant.Besides, it can also be used when T>N and when N>T.The test, which is based on VAR, assumes that there is no cross-sectional dependency.(Akbaş, 2013:791-812) We investigate the causal relationship among the 'exrate' 'intreserves' 'moneysupply' are cointegrated and BFCIUS for Fragile Five countries between 2000:01 and 2017:01.This study is a causality analysis in which we apply the Dumitrescu Hurlin Panel causality method.This third approach proposed by Elena Ivona Dumitrescu and Hurlin (2012) does account for the cross-sectional dependence.Dumitrescu Hurlin Panel Causality approaches have been employed to examine the direction of causality in the panel data.According to the significance level; for 5%; (*) for 1% (**)Lag length selection based on Scwarz Info Criterion with a max lag of 8.
As seen from the Table 4 and Figure 2 it was displayed that the variable of the BFCIUS is a homogeneously cause of the international reserves and exchange rates.It was observed that there is bidirectional causality between the money supply and exchange rate.The reason of the change in the international reserves was classified under three variables: the money supply, exchange rate and BFCIUS.Thus, the findings displayed that from the perspective of panel causality results a variation in the international reserves carries the feature of being a result, whereas the variables in the BFCIUS, exchange rate and money supply carries the feature of being a cause.From this point of view it was defined that on account of Fragile Five countries the BFCIUS is a significant variable for the exchange rate and international reserves.
We use a panel-data vector autoregression methodology.This technique combines the traditional VAR approach, which treats all the variables in the system as endogenous, with the panel-data approach, which allows for unobserved individual heterogeneity.We specify VAR model as follows: The most important issue on the implementation of VAR model is to select the proper lag length.A VAR with m variables, all m variables should be constant.In this study, for the selection of proper delay for estimated models, optimal delay was defined as "6" with SIC and the results of delay selection criteria were listed in Appendix 2. The impulse coming to the BFCIUS causes large volatility on the exchange rate and this volatility is seen from the impulse response diagram stabilized in the 9 th period.Consequently, relatively large response on the exchange rate created by the BFCIUS was observed.

Impulse Response Analysis
It is understood that the impulse of money supply generates very low response in the exchange rate and this response can easily be stabilized.The impulse in the international reserves causes high response in comparison with the money supply, but low response in comparison with the BFCIUS and it was displayed that this response firstly causes positive fluctuation on the general trend, but from the 5 th period causes negative fluctuation.
As it can be seen from the Roots of Characteristic Polynomial all of the modulus roots are smaller then one and also all roos are inside the unit circle which means system is stationary.

Variance Decomposition Analysis
Interpretation of individual factors in the estimated VAR models is difficult; therefore, interpretation with impulse-response analysis and variance decomposition methods is required.The direction and level of reaction of variables in five different VAR models to the shocks in action-reaction analysis and error terms are displayed in Graph 1.In impulse-response analysis and variance decomposition, the order of variables entering the calculation is important.The order of the variables should be from exogenous to endogenous.Based on Granger causality results and theory, the series were added to VAR Analysis model from exogenous to endogenous in order of; 'blmindex', 'exrate' 'intreserves' and 'moneysupply' When the findings of the Variance Decomposition given in the Appendix 5 were examined, it was defined that the variable in the exchange rate can be explained maximum with the variable in the BFCIUS.Additionally, it was determined the effects of this variable continued in a constant manner (3%) from the 3 rd period.The variable caused the change in the variable of the international reserve is again the BFCIUS and displaying its effects from the 2 nd period and continuing with relatively slight increase (5%) till the 10 th period are among the variance decomposition findings.

Conclusion
In this study it was aimed to display the basic dynamics of the value loss of the local currency of the Fragile Five countries especially against the dollar in the last years.For this purpose, the stability of variables was tested with the panel unit root tests in order to include to the panel data analysis and it was determined that except the BFCIUS the variables of the money supply, exchange rate and international reserves are non-stationary, whereas at the first difference they remain stability.Since in the level value the variable of the BFCIUS is stationary it was not included to The Panel Co-integration Test.At first differences the variables of the stationary money supply, exchange rate and international reserves were identified as co-integrated according to the results of the Pedroni Residual Cointegration Test.In other words, it was figured out that for the Fragile Five countries mentioned variables have association in the long run.The mentioned variables were examined from the perspective of the causality directions with the help of the Dumitrescu Hurlin Panel Causality Test.According to the results of Dumitrescu Hurlin Panel Causality Test it was identified that the BFCIUS is a 'homogeneously cause' of the international reserves and exchange rate.However, due to the other findings, the exchange rate is a 'homogeneously cause' of the international reserves and money supply.Also it was determined that the money supply is a 'homogeneously cause' of the international reserves and exchange rate.Thus, it was emerged the bidirectional causality between the exchange rate and money supply.According to the findings of the Panel CausalityTest results, impulse response and variance decomposition were interpreted, as well as it was estimated with the Panel Vector Autoregression Model (PVAR). As The studies on the effects of Central Bank interventions on exchange rate volatilities have quite complex empirical findings.On the other hand, common results of these studies indicate that, such central bank interventions have effective yet short term effects on the exchange rate levels.Under flexible exchange rate regimes, central bank interventions result in increased volatilities in exchange rates (Ishii et al. 2006:1).In their analysis, Adrangi et al.(

Figure 1 .
Figure 1.Currency Composition of Ofiicial Foreign Exchange Reserves % casuality techniques were used to uncover the direction of relation beween variables.Thereby, Panel VAR (PVAR) model 62the variables.The exchange rate index, data set of which is provided by Bloomberg, involves the variables of international reserve index, money supply index and BFCIUS; and the monthly data comprise of 1015 observations involving March 2000-January 2017 period.

Table 1 BLOOMBERG
Source.This table lists the major sub-indexes and their underlying indicators that form the Bloomberg Financial Conditions Index US.The table is modeled after Rosenberg(2009), pp.11

Graph of Exchange Rate, Money Supply, Internationa Reserves and for Fragile Five Countries
Pedroni (1999)cs proposed by Levin, Lin & Chu (2002), Im, Pesaran and Shin (2003), ADF Fisher and Fisher-type unit root test tests using ADF and PP tests proposed by Maddala and Wu (1999) are employed.Second, we test for cointegration data employing the heterogeneous panel cointegration test developed byPedroni (1999)which allows different individual effects crosssectional interdependency.Third, we apply Panel Vector autoregression (PVAR) Model to Fragile Five countries panel data from 5 countries to study the dynamic relationship between exchange rate and international reserves, money supply and BFCIUS.
exrateit; exchange rate(USD/Local Currency) intreservesit; International reserves (Millions / USD) moneysupplyit; Money Supply (Local Currency) blmindexit; The Bloomberg U.S. Financial Conditions Index Detailed definitions of the variables are given in Appendix 1. Graph 1.

Table 3 . Pedroni Residual Cointegration Test
According to the test statistic indicates that the relevant variables stationary.Lag length selection based on Scwarz Info Criterion with a max lag of 14.According the Pedroni Residual Cointegration Test Result we reject the null hypothesis rather we accept the alternative hypothesis both under %5 and %1 significance level.So we concluded that 'exrate' 'intreserves' 'moneysupply' are cointegrated and they have a long run relation.

Table 4 . Dumitrescu-Hurlin Panel Granger Causality Test Results
indicated by the analysis results, there are significant and asymmetric relationships among the depreciation values of the local currencies of Fragile Five Countries against USD in 2000:03 -2017:01 term.Local currency depreciation values against USD are as follows: USD/Brazilian Real Parity 81,37%, India USD/Indian Rupee Parity 55,73%, South Africa USD/South African Rand Parity 107,69% and Turkey USD/Turkish Lira Parity 533,33%.