Understanding the common dynamics of the emerging market currencies

Economic Modelling 49 (2015) 120–136

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

Understanding the common dynamics of the emerging market currencies☆ Meltem Gülenay Chadwick a,⁎, Fatih Fazilet b, Necati Tekatli c a b c

Central Bank of Turkey, Research and Monetary Policy Department, Istiklal Caddesi 10, Ulus, 06100 Ankara, Turkey University of Minnesota, Department of Economics, Minneapolis, MN 55455, United States Istanbul School of Central Banking, Central Bank of Turkey, Atlihan Sk 30A, Kadikoy, 34726 Istanbul, Turkey

a r t i c l e

i n f o

Article history: Accepted 12 March 2015 Available online xxxx Keywords: Emerging market currencies Dynamic factor model Empirical exchange rate models Nonlinear models Factor-augmented vector autoregression

a b s t r a c t The aim of this study is twofold. First, we examine if there exists a common movement among the currencies of emerging markets that implemented flexible exchange rate regime after 2000. Second, we examine whether this comovement is closely related to financial market conditions and macroeconomic fundamentals in emerging market economies. Our findings suggest that currencies of the emerging market economies have a common movement, which we name as “Exchange Rate Index”. We find that the Exchange Rate Index can be explained to a great extent by financial market indicators while macroeconomic fundamentals have relatively less power in understanding this common exchange rate pattern. The results particularly underline the importance of sovereign debt risk, equity return differentials and risk appetite. The relationship between financial variables and the Exchange Rate Index is significantly nonlinear, while the results for macroeconomic fundamentals do not show any nonlinearity. © 2015 Elsevier B.V. All rights reserved.

1. Introduction After the recent global crisis, which strengthened the significance of global financial linkages, increased capital flows to and within emerging economies became a follow-up topic for policy makers and researchers, as these flows revived the agenda of macroeconomic risks. While developed countries were declaring quantitative easing measures, emerging countries started taking a number of preventive measures toward handling pressures on their exchange rates, protecting their competitiveness in global markets and keeping their current account movements stable. Within this study, we aim to document the commonalities in nominal exchange rates of emerging markets and investigate the driving forces of it, as the research on emerging market currencies is relatively scarce. Our study contributes to a body of research focusing on financial and macroeconomic background within the emerging market economies and their corresponding effects on exchange rate movements that became more prominent especially after the last global crisis.

☆ We thank the participants at the 20th Symposium of the Society for Nonlinear Dynamics and Econometrics, EuroConference 2011, 2012 Annual Conference of the Center for Economics and Econometrics and the seminar participants at the Central Bank of Turkey for their useful comments. We wish to thank the two anonymous referees for their valuable advices and comments that helped improve the paper. We would like to thank A. Bubula and I. Otker-Robe for sharing the “de facto” exchange rate regime data for emerging market countries. The views expressed herein are those of the authors and not necessarily those of the Central Bank of Turkey. ⁎ Corresponding author. E-mail address: [email protected] (M.G. Chadwick).

http://dx.doi.org/10.1016/j.econmod.2015.03.011 0264-9993/© 2015 Elsevier B.V. All rights reserved.

In the 2000s, there have been dramatic changes in the exchange rate policies of emerging countries and the number of countries which prefer more flexible currency regimes increased considerably when compared to the 1990s. As more emerging countries have experienced the flexible exchange rate regime within the recent years, the importance of understanding exchange rate dynamics in these countries has increased significantly for policymakers, economists and investors. In this paper, we first show that the dynamics of exchange rates in emerging markets show a common pattern and we extract this common pattern using a dynamic factor model. We introduce this common factor as a composite index as in Stock and Watson (1989) and call it as the “Exchange Rate Index” for emerging market economies. To our knowledge, the research on the common dynamics of exchange rates is a very intact area.1 The Exchange Rate Index that we extract explains a significant portion (nearly 60%) of the variation in currencies. In this way, the common dynamics can be followed as a fundamental pattern that will reflect the aggregated common currency movement for emerging countries and their common pattern can be studied in a unified manner. Next, we attempt to explain the dynamics of the Exchange Rate Index using the financial market conditions and macroeconomic 1 There exist a couple of papers that are related to the common movement of exchange rates. One of them is by Engel et al. (2009) about forecasting the exchange rates of 17 OECD countries, and constructing factors from a cross section of exchange rates. The papers by Cayen et al. (2010) and Kempa and Wilde (2011) investigate the similarities and differences in the real exchange rate dynamics of several developed economy currencies. Aggarwal and Simmons (2008) examine the common stochastic trend among Caribbean currencies, but this study gives information about comovement between two pairs of currencies.

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

fundamentals of emerging economies relative to the US as the world economy. To this end, we employ exchange rate models derived from the theory and also apply univariate or multivariate econometric analysis when necessary. Theoretical literature states that exchange rates are determined by fundamental variables. However, in international macroeconomics and finance literature, many econometric analyses show that there is a disconnection between exchange rate movements and macroeconomic fundamentals such as output, money supply, inflation and interest rates. The literature on the determinants of exchange rates have mainly focused on developed countries (Cayen et al., 2010; Engel et al., 2009; Kempa and Wilde, 2011; Wang and Wu, 2012). However, it has become widely accepted that empirical models using macroeconomic variables have notoriously limited success in explaining exchange rates for developed countries. Meese and Rogoff (1983), Obstfeld and Rogoff (2000) and Cheung et al. (2005) are among the leading papers showing this poor explanatory power of macroeconomic fundamentals. On the other hand, literature on the relationship between exchange rates and macroeconomic fundamentals are not all doom and gloom. Rapach and Wohar (2002) found support for a simple form of the long-run monetary model of exchange rate determination for a collection of 14 industrialized countries. Taylor and Peel (2000) also have supportive evidence for the relationship between exchange rates and macroeconomic fundamentals over the long run with respect to a nonlinear mean reversion toward the monetary fundamental equilibrium. We begin our analysis on the determinants of the Exchange Rate Index by focusing on financial markets first and extend the analysis to the macroeconomic fundamentals of emerging economies relative to the world economy (the US economy). The financial markets in these economies have gained more attention over time to investigate the determinants of exchange rates, as the implementation of more flexible regimes within the last ten years made it possible to analyze the effects of financial markets and capital flows on currency movements without any intervention effect in a healthy manner, and the topic gained its well-deserved prominence after the recent global crisis. In addition to that, there is a huge gap in the exchange rate literature about the employment of financial variables to apprehend the emerging market exchange rates.2 With this paper, we try to fill this gap showing that the Exchange Rate Index is highly linked to the sovereign debt risk, differentials of equity returns between the emerging markets and the US, and investor risk appetite. This relationship is strengthened considerably with the use of nonlinear estimation methods. We then employ exchange rate models built for macroeconomic fundamentals in the literature; that is, purchasing power parity model and sticky price model. Showing that both exchange rate models do not exhibit nonlinearity, we conduct a factor-augmented vector autoregression (FAVAR) model to study the implications of the purchasing power parity and sticky price assumptions in a multivariate manner. Similar to most papers in the literature, our preliminary analyses with macroeconomic fundamentals fail to explain the exchange rate dynamics. However, when univariate analysis is replaced with multivariate analysis and endogeneity is taken care of, both models can explain the Exchange Rate Index to some extent, though not as strong as financial variables, and the tests show no substantial nonlinearity between the macroeconomic fundamentals and the Exchange Rate Index. In Section 2, we introduce the exchange rate database used in our paper. In Section 3, we introduce a dynamic factor model with the aim of extracting the common movement of selected emerging market 2 Linkages between financial variables and exchange rates have been examined in economic literature by various papers. For example, Pan et al. (2007) summarize the theoretical and empirical literature on the relationship between exchange rates and stock prices in detail. However, there does not exist a clear empirical paper that finds a significant contemporaneous relationship between exchange rates and stock prices. Existing literature basically use stock prices as one of the main financial variables to explain the exchange rate movements. Unfortunately, this literature is rather weak for analyzing the building blocks between significant financial fundamentals and exchange rate dynamics.

121

currencies that we name Exchange Rate Index. In Section 4, we examine the link between the Exchange Rate Index and the financial variables. In Section 5, we run a similar analysis to that used in Section 4, only this time using macroeconomic fundamentals. The final section concludes. 2. Database of selected emerging market currencies Before investigating the common dynamics of emerging market currencies, it is essential to examine and understand the exchange rate regimes of the emerging countries in question. The reason for such an inquiry is that; during times of increased risk appetite, an emerging country with a flexible exchange rate regime can experience appreciation due to increased capital inflow. However, the same exchange rate movement will not be observed in an emerging country with a fixed exchange rate regime. Moreover, currency movements of an emerging country within fixed regimes can move in different ways, compared to other countries with flexible exchange rate regimes in case of increased risk appetite environment. Therefore, emerging countries that have been included in our study are chosen according to their exchange rate regime. Emerging countries with flexible exchange rate regimes are chosen for a healthy analysis. Determination of exchange rate regimes has been accomplished by using a “de facto” exchange rate classification scheme produced by the International Monetary Fund (IMF)3. The classification in question is composed of eight course classes where a number is assigned for each course. In this framework, number one refers to most rigid exchange rate regime, where the most flexible regime is represented by number eight, and the numbers in between one to eight represent exchange rate regimes with different ranges of rigidities.4 Before estimating and analyzing the common movements of emerging market currencies, we examine the exchange rates of thirty countries and chose fourteen among them that have relatively more flexible exchange rate regimes. Between the years 1990 and 2000, selected fourteen emerging market countries mediated their currency markets quiet often. Therefore, for the 1990–2000 period, it is hard to decompose the effects of intervention from currency movements for analyzing the effect of financial or macroeconomic fundamentals on the dynamics of exchange rate movements alone. Thus, we leave that period out of analysis when we were conducting our study. On the other hand, there have been slight interventions to exchange rate markets after the second quarter of 2009, during the time when emerging market countries were aiming to take some measures to control short term capital flows. This slight intervention is more observable within 2010, when the emerging market currencies started to appreciate with increased capital inflows. However, these interventions are not as strict as those in the 1990–2000 period. Fig. B.1 illustrates the evolution of the selected fourteen emerging countries' currencies after 2000. Nominal exchange rate data against the US Dollar is monthly averages from Bloomberg database starting from January 2000 and ending in August 2014. All the definitions for the series used in this research is at Table A.1. Table A.2 reports descriptive statistics of raw nominal exchange rates. The database used in Fig. B.1 has been standardized due to a significant scale difference between the emerging economies. 3. Common movement of emerging market currencies: Exchange Rate Index There is an extensive literature on the determinants of exchange rates,which has mainly focused on developed countries. However, the literature has scarce analysis on exchange rates of emerging economies 3

See Bubula and Ötker (2002) for details. The information was retroactively updated by A. Bubula and I. Otker-Robe, “The Evolution of Exchange Rate Regimes since 1990: Evidence from De Factor Policies,” IMF WP/02/155. The official definitions of the categories are available at: http://www.imf. org/external/np/mfd/er/index.asp. 4

122

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

and almost no analysis on the common dynamics of these exchange rates. In this paper, we conjecture that there is a common movement in the emerging market exchange rates, because there are a lot of commonalities between these economies that can affect the exchange rate dynamics similarly, such as exchange rate regimes, monetary policies, financial integration and capital flows. Along with the descriptive statistics given above, cross correlations of the exchange rates of the emerging market countries examined in this study take place in Table A.3. On average, correlations are high between most of the countries and they seem to be free from regional or any other grouping effects. Moreover, we can also test if there is a common pattern between these variables. To this end, we conduct a likelihood ratio test and reject the null that says no common pattern. Factor analysis is an econometric approach that can be used to analyze interrelationships between the currencies and to explain these variables in terms of their common underlying dimensions. It is a method for condensing the information contained in a larger set of variables into a smaller set of dimensions, while keeping loss of information to a minimum. To estimate the common component of exchange rates of selected emerging market countries, we use a dynamic factor model cast in a state space representation. We introduce this common factor extracted using the dynamic factor model as a composite index as in Stock and Watson (1989) and call it as the “Exchange Rate Index” for emerging market economies. Before estimating a dynamic factor model to extract the common dynamics of our nominal exchange rate database for selected emerging market countries, we need to test if there does exist a significant common pattern among emerging market currencies and if this common pattern can be represented by a common factor using a single factor model. To this end, first we conduct a likelihood ratio test with the null hypothesis that the number of common factors is zero. We reject the null hypothesis with a probability value of “0.000”. Second, we use an information criteria developed by Bai and Ng (2007) to determine if there is at least one factor that can represent the common dynamics. The results show that the variation of selected currencies can be summarized by at least one factor. The factor model adopted in this study tries to formalize the idea that the Exchange Rate Index (ERI) is a representative “Reference Cycle” or a “Coincident Index” which is best measured by looking at comovements across several aggregate time series.5 In this study, we follow the methodology proposed by Giannone et al. (2008) who develop a parametric dynamic factor model cast in a state space representation and estimate the factors in two steps.6 Usage of factor models to extract the common fluctuations for some macroeconomic fundamentals is not new in the economic literature. Smith and Zoega (2007) use a factor model to extract the global component of 21 OECD countries. Kose et al. (2003) utilize dynamic factor models to extract the common components among different macroeconomic variables within the framework of business cycles employing Bayesian techniques. In addition to Kose et al. (2003, 2008), Neely and Rapach (2011) also use a dynamic factor model to analyze the comovement of cross country inflation rates. With this study we apply a dynamic factor model to nominal exchange rates of selected emerging countries as follows; let Xt = (x1,t, x2,t, …, xn,t)′ denote the monthly exchange rate series of 14 emerging market countries; hence n = 14. We assume that Xt has the following factor model representation: X t ¼ μ þ Λ f t þ εt

ð1Þ

where, ft is a 1 × 1 unobserved component, which represents the common dynamics of the emerging countries, and εt is a vector of 5

See Stock and Watson (1989) for details. Details of the model and estimation process take place in Doz et al. (2011) and Banbura et al. (2010). 6

idiosyncratic components that are assumed to have zero mean. Hence, the constants μ = (μ1, μ2, …, μn)′ are the unconditional means. Further, the common factor is modeled as an autoregressive process of order one: f t ¼ Θf t−1 þ ut ;

ut ∼ i:i:d:N ð0; Q Þ

ð2Þ

where Θ is the k × k matrix of autoregressive coefficients. Finally, we assume that the idiosyncratic component follows an AR(1) process: εi;t ¼ α i ε i;t−1 þ ϑi;t ;

  2 ϑi;t ∼ i:i:d:N 0; σ i

ð3Þ

with E[ϑi,tϑj,s] = 0 for i ≠ j. Doz et al. (2006) have shown that, for large cross-sections, the model given by Eq. (1) can be estimated by maximum likelihood under the assumption of lack of cross-sectional correlation in the idiosyncratic component.7 Using the dynamic factor model structure defined above we extract the unobserved factor, which explains nearly 60% of the variation of the emerging market currencies, out of total variation. If we exclude the least correlated four countries, this number rises to 73%. In the remaining parts of this study, we report and analyze only one factor which we introduce as the Exchange Rate Index, which we will denote as “ft”, for emerging market economies.8 The evolution of this index as univariate time series, which represents a significant part of the emerging market exchange rate dynamics, can be observed via Fig. B.2. As can be observed in Fig. B.2, the Exchange Rate Index moves parallel to the general trend of country exchange rates within the period of 2000–2014. This figure is a good indication of how well the index represents the common movement of emerging market currencies in a clear way.9 If the movements of index and the country exchange rates are examined, we can state that effects of precautionary measures against the speculative capital inflows can be clearly observed toward the second half of 2010.10 Table A.4 shows the variance explained by the common factor for each of the emerging market countries. During certain times, major economic and financial events can be linked to the common dynamics of the currencies strongly due to both the commonalities among these economies and the significance of these events for most emerging markets. Before we start the analysis to observe the impact of macro-financial variables, we can graphically illustrate the links between some of the possible major events and turning points of the currencies due to these events via Fig. B.3. As can be seen in Fig. B.1, after year 2000 emerging market currencies depreciated first due to the 2001 crisis, then switched to a trend of consistent appreciation with the rise of the emerging market countries and the historical low interest rates in the US. In 2008, the deepening effects of the global financial crisis resulted in depreciation and finally currencies appreciated again with short term capital flows to emerging markets after the Quantitative Easing employed by the US Fed in 2009. In 2010, sovereign debt crisis in the Euro area led to rise in the currencies. In 2013, there is a slight depreciation due to the effects of exit strategies in the US by the Federal Reserve. We will further study the linkages between the macro-financial events and the movements in emerging market exchange rates represented by the log difference of Exchange Rate Index within the next 7 Details of the maximum likelihood implemented by Expectation Maximization (EM) algorithm can be found in Banbura et al. (2010). 8 Robustness of the model results has been tested by three different dynamic factor models estimated using a Bayesian algorithm, and results do not differ qualitatively. Details of these models take place in Tekatli (2010), Chadwick (2010) and Kim and Nelson (1998). 9 Also please see Fig. B.4 and Fig. B.5 that illustrate the Exchange Rate Index with individual nominal exchange rates of 14 countries. 10 How accurate the exchange rate represents the emerging market currencies can be observed more clearly when we plot the exchange rate index with each country currency. You can see those plots at Appendix B.4 and B.5.

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

chapters. Yet, with a quick examination of Fig. B.3 one can follow the movements of the change in the Exchange Rate Index and the timing of the events, some of which are already discussed in this section. Fig. B.3 is indeed very self-explanatory to observe and understand how the log difference of the Exchange Rate Index is intuitively linked to the major economic events. We also conduct general causality tests for the data we use to explain the Exchange Rate Index to see whether there exist any causal relation between the financial and macroeconomic variables and the Exchange Rate Index. Granger causality tests will help us understand if the “known events of the period” that can be proxied by the dataset we use may or may not cause such a relationship. Indeed, such a causality analysis is also important as we use various models where the Exchange Rate Index is being explained successfully by financial variables and partially explained by macroeconomic variables. In Tables A.5 and A.6, we present the results of Granger causality analysis that will give some idea about the links between macroeconomic and financial dynamics that can be proxied to some extent by the data utilized and the Exchange Rate Index.

4. Financial variables and the Exchange Rate Index In the previous section, we point out that the Exchange Rate Index that has been extracted from the dynamic factor model represents approximately 60% of the total exchange rate movements of the selected emerging market countries. In this section, we will examine the relationship between the Exchange Rate Index and selected financial variables. There are two reasons why we give special emphasis to explaining exchange rate dynamics using financial variables. The first reason is due to the fact that the implementation of more flexible regimes within the last ten years made it possible to analyze the effect of financial markets and capital flows on currency movements in a much healthier manner that can now be observed without any intervention effect. The second and more important reason is that there is a huge gap in the exchange rate literature about the treatment of financial variables to apprehend the movement of exchange rate dynamics, and the remaining parts of this section will illustrate how important it is to fill this gap. We use different financial variables related to different types of asset markets in this study. The first variable represents the stock market i.e.; S&P500 index and MSCI emerging market stock price indices which reflect the developments in equity markets of the emerging countries. The second variable represents bond market and the variable used is EMBI+ Sovereign Spread emerging market main index, which can be considered as a risk indicator for the bond markets of emerging market countries. Finally, we use a variable that can represent the global economic and financial risk conditions; VIX index that is also considered as a good indicator of risk appetite. Static and dynamic correlations of these financial variables and the Exchange Rate Index, are illustrated in Table A.7.11 In Table A.7, while negative correlations are related with the equity returns, positive correlations refer to indicators related to risk. The signs of the correlation coefficients are in line with the expectations. The correlations are at their highest values in the medium term, which implies that in a couple of months these variables move further in the same direction in absolute terms. In the short-, medium- and long-term, the equity market indices (mainly, MSCI indices) are the ones that are highly correlated with the Exchange Rate Index (the coefficients range from − 0.71 to − 0.91). On the other hand, the Exchange Rate Index and bond market indices have a correlation 11 We use the definition of dynamic correlation from Croux et al. (2001) for bivariate time series analysis, to analyze the covariance matrices between exchange rates and financial series in frequency domain, instead of the time domain. This approach describes the dynamic properties of comovements between two series through their frequency spectrum. The advantage of dynamic correlation is to measure the strength of the relationship between two time-series within different horizons.

123

coefficient changing from around 0.40s to 0.70s. Therefore, this correlation analysis provides the empirical evidence that the Exchange Rate Index representing the emerging market currency dynamics is strongly related to financial variables and financial indices for bond and stock markets. In order to understand if financial conditions can explain the dynamics of the Exchange Rate Index for emerging economies, we study each financial variable in a separate manner. We start analyzing the link between Exchange Rate Index and financial variables with equity returns. We will consider the relationship between Exchange Rate Index and equity market indicator, which is the MSCI emerging market index, separately as these two variables may be endogenous.12 Fig. B.6 shows the relation between the Exchange Rate Index and the MSCI emerging market return index in a clear manner. We analyze the relationship between Exchange Rate Index and equity returns using the “Uncovered Equity Return Parity” (UERP) model explained below. To illustrate the dynamics of this relation in a much healthier manner we used an estimation technique which is robust to the endogeneity problem, i.e. Generalized Method of Moments (GMM) of Hansen (1982).13 The UERP model can be explained briefly as: no arbitrage condition implies that when expected equity returns in a country/region are lower than expected equity returns in another country/region, the currency associated with the market offering lower returns is expected to appreciate (see also Cappiello and Santis (2005)). Accordingly, the UERP model can be represented as:   US Δf t ¼ γ þ α r t −r t þ ut

ð4Þ

The UERP model tries to explain the percentage change in exchange rates with excess equity returns. In Eq. (4), Δft represents the log difference of the Exchange Rate Index representing the common movement of selected emerging market countries and rt refers to equity returns. Results of the estimated model illustrated in Eq. (4), is reported in Table A.8.14 Table A.8 illustrates the estimated UERP model. UERP model is based on market efficiency and risk neutrality assumptions. Therefore, we expect coefficient γ to be statistically insignificant and coefficient α to be minus one.15 When the results are examined through Table A.8, the constant being insignificant signals the market efficiency meaning that the theory underlying the model is met. Negative coefficient attached to the excess equity return illustrates the fact that when the equity returns are relatively higher, exchange rates appreciate to compensate for it, which is in line with the theory. However, the coefficient attached to excess equity returns being smaller than one in absolute terms signals to a risk premium that is not captured by the model. On the other hand, the last row of Table A.8 underlines the strong relation between exchange rates and the excess equity returns, as the direction of change in both exchange rate and equity returns matches intuitively by 68% and are consistent with the expectations.16 Estimation results show that 12 Theory behind the relationship between exchange rates and stock prices is not well built and clear as the causality between the two remains an open-ended question. Dornbusch and Fischer (1980) suggest that changes in exchange rates affect the competitiveness of multinational firms and hence it affects the stock prices. Accordingly the causality is from exchange rates to stock prices. On the other hand, according to the portfolio balance approach, exchange rates are determined through a market mechanism and when the stock market is on the rise, it will attract capital flows from foreign investors which can increase the demand for the country's currency. Accordingly, the causality is from stock markets to exchange rates. 13 The details about estimated model and “Uncovered Equity Return Parity Condition” take place in Cappiello and Santis (2005). 14 In this paper we use Δft, which is extracted from the log difference of standardized nominal exchange rates using the dynamic factor model explained above, when we examine the driving factors of the exchange rate index. 15 This would be in line with what the UERP model predicts to arbitrage away any profit possibilities. Extended discussion can be found in Cappiello and Santis (2005). 16 The R2 of similar estimations in the literature appears to be much lower than what we find after estimating Eq. (4).

124

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

there exists a risk premium that we cannot price with these simple models, which are priced by agents in the market. These results are very essential for understanding the relationship between emerging market countries' exchange rates and excess equity returns. The current financial crisis has caused sharp imbalances in global exchange rate structures and some other financial fundamentals, which clearly demonstrates the financial frailties of countries on a global scale. With the recent crisis, collapse in asset prices and underlying capital flows, flight-to-safety phenomenon gained a special emphasis where both the US and other investors have shifted from equities into fixed income instruments, particularly to safe government bonds. After some European countries announced the big debt ratios and default risk they face, European governments and the markets struggled to find a solution to possible crisis due to government debt. Therefore, it is important to understand how global risk conditions influence currency dynamics. Accordingly, risk exposures of emerging market countries became very vital, and we use two series that carry necessary information about risk exposures, EMBI+ Sovereign Spreads and VIX.17 Figs. B.7 and B.8 demonstrate the strong correlation between the Exchange Rate Index, in a visual way. Hence, first we perform least squares to examine the explanatory power of EMBI + Sovereign Spreads and VIX on Exchange Rate Index of the selected emerging market countries. Linear regression results of the Exchange Rate Index on the financial variables are demonstrated in Table A.9. Table A.9 shows us that, when there is an increase in VIX or sovereign bond spreads of emerging market countries, which is the situation referred to as the increased financial risks in those countries, resulting in capital outflows from these countries and capital outflows from emerging countries increase the demand for the global currency (the US dollar), but at the same time increases the supply of currencies in question. Therefore, EMBI + Sovereign Spreads and VIX have a positive relation with the Exchange Rate Index. Figs. B.7 and B.8 also hint that sudden big changes in the financial variables affect the Exchange Rate Index relatively more compared to small changes. In other words, when there is a significant change in the risk environment, the Exchange Rate Index can be more sensitive to changes in financial variables.18 This observation of the sensitivity of the Exchange Rate Index can refer to a situation where the relation between the Exchange Rate Index and the financial variables in question being nonlinear, and therefore, we will examine the explanatory power of EMBI+ Sovereign Spreads and VIX on the Exchange Rate Index via two different nonlinear models; i.e. a threshold model and a Markov switching model.19 The first model that we employ is the threshold model of Hansen (2000).20 The threshold model used in this study can be illustrated as: Δf t ¼ β1′ xt þ et ; Δf t ¼ β2′ xt þ et ;

qt ≤γ qt Nγ

ð5Þ

where Δft represents the Exchange Rate Index of emerging countries extracted from the dynamic factor model using logarithmic difference of standardized nominal exchange rates, xt represents the financial 17 There exists a literature using the asset pricing approach to explain foreign exchange movements in terms of compensation for risk. Details can be found in papers such as; Fama (1984), Hodrick (1989), Dumas and Solnik (1995) and a recent one Adrian et al. (2009). 18 For reference and details please see Cairns et al. (2007). 19 The linkage between financial variables and exchange rates has been the subject of various papers within the framework of nonlinear models. One such paper is by Walid et al. (2011) that examines the dynamic linkage between stock price volatility and exchange rate changes for four emerging countries using a Markov switching EGARCH model. Fiess and Shankar (2009) analyze the importance of financial fundamentals in the central bank exchange rate policy applying regime switching methods. Finally, Ang and Timmermann (2012) summarize the regime switching models in relation with the financial market in a very intuitive way. 20 We confirm the nonlinearity of the relationship using the nonlinearity test offered by Hansen (2000). We show the test results in Fig. B.9.

variable (either VIX or EMBI+ Sovereign Spreads) and qt is the threshold variable used in the model.21 Absolute value of the change in the Exchange Rate Index has been used as the threshold variable in this model. Using this threshold variable, the sample has been divided into two regimes, where first regime represents the small changes in the Exchange Rate Index and second regime represents the big changes in the Exchange Rate Index. On the other hand, the threshold value which is important for deciding on two regimes has been chosen endogenously within the model estimation procedure. Estimated threshold values take place in Table A.9. The results of the threshold regression have been illustrated in Table A.9. Bold numbers in Table A.9 refer to statistically significant variables, and results demonstrate that the threshold regression results are more significant, especially for the second regime (where the threshold is bigger than 3.010) when compared to Global OLS results.22 If the results from the threshold model are examined closely, second regime which refers to the sample with higher values than the threshold value, both VIX and EMBI + Sovereign Spreads have a higher explanatory power, and also this regime includes most of the variation in the whole sample. These results point out that when there is a significant change in the global risk appetite, or the risk exposure of the emerging market countries, i.e. their currencies depreciate or appreciate accordingly in relation to the risk conditions. Exchange rates are characterized by highly persistent patterns, punctuated by abrupt changes, in which regime switching models capture well.23 Therefore the second model we consider is the following regression model with Markov-switching parameters: Δ f t ¼ βSt xt þ et   2 et ∼ N 0; σ St βSt ¼ β0 ð1−St Þ þ β1 St 2

2

2

σ St ¼ σ 0 ð1−St Þ þ σ 1 St Pr½St ¼ 1jSt−1 ¼ 1 ¼ p11 Pr½St ¼ 0jSt−1 ¼ 0 ¼ p00 where xt (in our case it is either the VIX or EMBI+ Sovereign Spreads) is exogenous or predetermined and conditional on St − 1, and St is independent of xt.24 The results are illustrated in Table A.10. As seen in Table A.10 all of the coefficients are statistically significant in both states. The sign of the constant changes for different states and the relationship between the Exchange Rate Index and the financial variables are stronger for the first state. For the independent variable VIX, the first state is where the variance is lower, on the other hand for EMBI+ Sovereign Spreads variable the first state refers to a higher variance. To understand the threshold cut-off points and its relation to the significant economic developments, one can study Fig. B.3. As a robustness, we repeat all the steps till this part of the paper, i.e. we extracted the Exchange Rate Index and estimated nonlinear models, using two separate subset of countries. We selected two subsets of countries within the framework of two criteria: i. possible highest variation explained by the Exchange Rate Index without losing the focus of the paper – that is creating a nominal Exchange Rate Index from leading emerging market currencies; ii. the highest possible average cross correlation between currencies within the selected group of countries. The first set of countries (Group 1) happen to be Brazil, Chile, Colombia, Czech Republic, Mexico, Philippines, Poland and Thailand, which has a variation explained above 83%. The second set of countries (Group 2) are ended up as India, Indonesia, Korea, Romania, South Africa and Turkey, which has low variation explained lower than 21

See also Franses and van Dijk (2000). We would like to state that all the estimations (OLS and threshold model) are done with transformed variables. We take the log difference of the variables used to make them stationary. 23 See Ang and Timmermann (2012) for details. 24 See Kim and Nelson (2003) for the likelihood function and other details. 22

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

18%. Then, we conducted the analysis that take place in this section with these two groups of countries to see if the financial variables become more explanatory for the new Exchange Rate Index within different groups. From the model estimations conducted for both sets of Exchange Rate Indices, we concluded that the results within the subsets of two different groups of countries do not differ much from the current analysis.25 5. Exchange Rate Index and macroeconomic fundamentals In the previous part, we showed that the Exchange Rate Index which we constructed using the nominal exchange rates of 14 emerging market countries, can be explained significantly by indices related to equity prices (MSCI), volatility index related with the stock market (VIX) and indices related with bond market (EMBI + Sovereign Spreads). In this part, we will focus on the most prominent empirical exchange rate models in the economic literature used to explain exchange rate dynamics, but we will do it from a different perspective relative to existing literature. With the Purchasing Power Parity model (PPP model) being the cornerstone of international finance literature, and the Sticky Price Monetary model (SPM model) being the most widely used structural model in the literature, they are most commonly referred and cited exchange rate models. However, as mentioned in the introduction, they do not have good score in explaining exchange rate dynamics. In this paper, we want to estimate and see if both PPP and SPM models have any significant explanatory power for the dynamics of the Exchange Rate Index of selected emerging market countries. To do so, as in the previous section, first we will estimate these two models as explained in the literature using OLS,26 and then we will test if there is any nonlinearity.27 After rejecting nonlinearity, we will continue with the FAVAR. The PPP model is one of the most well accepted theories about exchange rate movements, as it is commonly used to measure the degree of exchange rate misalignment. According to the PPP theory, a certain amount of money should have the same purchasing power in two different countries to prevent arbitrage opportunity. Hence exchange rate movement reflects the changes in price levels in both countries. The theory can be formally stated as: Δf t ¼ β0 þ β1 P^ t

ð6Þ

125

The SPM model29 can be interpreted as an extension of the PPP model, where the price variable is replaced by some macroeconomic fundamentals that capture money demand and overshooting effects. The SPM model can be formally stated as: ^ t þ ut ^ t þ β2 y ^t þ β3^it þ β4 π Δf t ¼ β1 m

ð7Þ

where mt is the logarithm of money, yt is the logarithm of real output, it is the interest rate, πt is the inflation rate and ut is the error term. “^” refers to the difference between emerging country macroeconomic fundamentals and US macroeconomic fundamentals.30 β1 and β4 are expected to be positive, since an increase in money supply or inflation leads to depreciation of money. On the other hand, the coefficients of yt and it are expected to be negative, since an increase in output leads to a decrease in real money balances, and an increase in it leads to an appreciation because of Uncovered Interest Rate Parity condition. Table A.11 illustrates the results of typical SPM model estimated in logarithmic differences, and from the global OLS estimates we clearly observe that linear models have little explanatory power, with only inflation being statistically significant. On the other hand, the threshold model cannot find any significant nonlinearity with only high explanatory power for the second regime. However, this regime has only seven data points.31 Fig. B.3 gives the picture of major economic turnovers so that one can compare the threshold cut-off points with the significant economic changes. We include the Exchange Rate Index in a VAR model together with macroeconomic fundamentals used in PPP and SPM models. In this analysis, we can capture the dynamics and the endogeneity between the interest rates and the exchange rates of the emerging market economies. Also the impulse responses allow us to study the interaction of the Exchange Rate Index and the fundamentals.32 We therefore estimate a VAR(2) for PPP model and a VAR(1) for the SPM model33: Results for the PPP model are illustrated with Eq. (8), which has much better explanatory power, and the shape of the impulse responses are in line with what the model predicts.34 A shock to price level causes depreciation and then the currencies start to appreciate again.35 Δ f t ¼ −0:19 þ 0:56 Δ f t−1 − 0:23 Δf t−2 þ 0:06 P^ t−1 þ 0:56 P^ t−2 ð0:25Þ

ð0:08Þ

ð0:08Þ

ð0:62Þ

ð0:58Þ

where in our setup Δft is the Exchange Rate Index of emerging markets, P^ denotes the difference of emerging countries' price level from the US

P^ t ¼ 0:19 þ 0:45 P^ t−1 − 0:12 P^ t−2 þ 0:05 Δ f t−1 − 0:01 Δf t−2

price level. The estimation results of the model are conducted using all the variables in log differences. In Table A.11 the OLS results for Eq. (6) according to the PPP model are given, as it is used in the exchange rate literature. Although the price variable is significant, it has got hardly any explanatory power to understand the exchange rate dynamics. We use threshold model to test for nonlinearity and reject nonlinearity significantly.28

LL ¼ −377:348

t

25

The results of the analysis for Group 1 and Group 2 can be provided upon request. Cheung et al. (2005) give an excellent review on these two models. In this paper we use the same notation as in Cheung et al. (2005) for PPP and SPM models. 27 Explaining exchange rates with macroeconomic fundamentals is becoming very popular within the framework of nonlinear models. Accordingly, Meese and Rose (1991) claim that the poor explanatory power of the macroeconomic fundamentals for exchange rates cannot be attributed to non-linearities for major OECD countries. Taylor and Peel (2000) suggest that the monetary model may possibly be rehabilitated when considered as a long-run equilibrium condition, where the adjustment toward this equilibrium may be nonlinear for the dollar against sterling and the mark. Frömmel et al. (2005) and Chen and Lee (2006) find evidence of a non-linear relationship between exchange rates and underlying fundamentals within the framework of Markov switching model. Yuan (2011) considers modeling the effects of the macroeconomic determinants on the nominal exchange rate to be channeled through the transition probabilities in a Markovian process and finds that the volatility of the exchange rate is associated with significant ARCH effects which are subject to regime changes. Finally, Kal (2011) finds that the relationship between economic fundamentals and the nominal exchange rates varies depending on the overvaluation or undervaluation of the currencies for four currencies. 28 Threshold model test results related for the nonlinearity is in Fig. B.10. 26

ð0:03Þ

ð0:08Þ

ð0:08Þ

ð0:01Þ

ð0:01Þ

ð8Þ

The FAVAR results for the SPM model are given with Eq. (9). We use VAR(1) with only the interest rate being endogenous. The Exchange Rate Index is better explained with this model where the impulse responses show that a shock to Exchange Rate Index causes a fall in interest rates.36 In summary, the FAVAR has been proven much useful than a simple OLS model to analyze the interrelation between the macroeconomic fundamentals and the Exchange Rate Index. With the exception of output, all of the fundamentals are significant in explaining the Exchange Rate Index.

29

See Dornbusch (1976) and Frankel (1979) for details. See Chinn and Alquist (2006) for a detailed analysis of SPM models. 31 Threshold model test results related to the nonlinearity is in Fig. B.10. 32 To do so we will conduct a FAVAR in first differences. See Bernanke et al. (2005) for details. 33 We choose the lag length according to AIC and BIC. 34 See Fig. B.11. 35 Values in parenthesis for Eqs. (8) and (9) are the standard errors. 36 See Fig. B.12. 30

126

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

^ t − 0:07 y ^ t þ 0:93 π ^t Δ f t ¼ −0:08 þ 0:44 Δf t−1 − 0:10 ^it−1 þ 0:22 m ð0:17Þ

ð0:07Þ

ð0:50ÞÞ

ð0:11Þ

ð0:36Þ

ð0:14Þ

^i ¼ −0:01 þ 0:26 ^i ^ ^ ^ t t−1 þ 0:01 Δf t−1 − 0:01 mt þ 0:01 π t − 0:004 yt ð0:02Þ

ð0:08Þ

ð0:01Þ

ð0:02Þ

ð0:06Þ

LL ¼ −415:239

ð0:02Þ

ð9Þ

As is done in the previous section for the robustness analysis, we again divide the countries into two (Group 1 and Group 2), i.e. ones with a higher total variation and higher cross correlation, and the ones with lower variation and lower cross correlation. We extract a common factor from these two sets of countries. From the results of the analysis for these two Exchange Rate Indices and two subsets of countries, we find that the main qualitative results of the paper do not change significantly.37 6. Conclusion In this study, we have shown that the exchange rates of selected fourteen emerging market economies, which practice flexible currency regimes, can be represented by a common factor that we name the “Exchange Rate Index”. Accordingly, this common factor can be followed as a time series index that represents approximately 60% of the total variation of emerging markets' currency dynamics. When the relationship between the Exchange Rate Index and financial variables

is investigated, we find that the financial variables have a significant explanatory power in currency movements of emerging countries in question. In particular, usage of nonlinear estimation techniques strengthens this result, both for the whole sample and the periods when there are sharp changes in the Exchange Rate Index. When we run the analysis using macroeconomic fundamentals within the framework of most popular empirical exchange rate models, we observe that macroeconomic fundamentals have hardly any explanatory power in explaining the Exchange Rate Index within the framework of OLS. In addition to that, the relationship between the Exchange Rate Index and macroeconomic fundamentals does not show any nonlinearity. On the other hand, using a FAVAR model improves the linkage between the Exchange Rate Index and macroeconomic fundamentals where we get reasonable impulse responses and most variables become significant. This paper underlines the importance of bond and stock market variables relative to the US as well as the risk measures to study the exchange rate movements of emerging countries. Specifically, when there is a significant change in the risk environment, the effects of this change on exchange rates are more pronounced and clearly observable. Therefore, as we have experienced in the near past, any precautionary measures taken by developed countries, such as liquidity management, might have an important effect on the exchange rate dynamics.

Appendix A. Tables

Table A.1 Data definitions. Data

Source

Period

Nominal exchange rates against USD VIX S&P500 MSCI emerging markets MSCI emerging Asia MSCI emerging Europe MSCI Latin EMBI + SS Main EMBI + SS Asia EMBI + SS Europe EMBI + SS Latin Industrial production CPI Index M1 Interest rates

Bloomberg Bloomberg Bloomberg Bloomberg Bloomberg Bloomberg Bloomberg Bloomberg Bloomberg Bloomberg Bloomberg IMF-IFS IMF-IFS IMF-IFS IMF-IFS

2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M08 2000 M01–2014 M06 2000 M01–2014 M06 2000 M01–2014 M06 2000 M01–2014 M06

Table A.2 Nominal exchange rates — descriptive statistics.

Brazil Chile Colombia Czech. Rep. India Indonesia Rep. Of Korea Mexico Philippines Poland Romania South Africa Thailand Turkey

37

Mean

Median

Maximum

Minimum

Std. dev.

Skewness

Kurtosis

Jarque–Bera

Probability

2.220 559.512 2181.081 24.243 47.736 9483.983 1123.208 11.431 48.010 3.349 2.992 7.943 36.507 1.472

2.151 538.530 2152.385 21.141 46.408 9203.000 1126.340 11.116 47.696 3.204 3.078 7.619 35.338 1.483

3.793 746.180 2959.300 41.286 63.754 12,162.000 1453.230 14.646 56.336 4.642 3.706 11.684 45.618 2.220

1.562 442.250 1733.280 14.938 39.368 7263.000 915.200 9.063 40.451 2.066 1.833 5.727 29.075 0.546

0.485 73.540 329.663 7.069 5.305 943.891 116.899 1.412 4.932 0.560 0.393 1.432 4.851 0.341

0.999 0.811 0.653 1.017 1.295 1.008 0.214 0.084 0.210 0.225 −0.868 0.837 0.220 −0.608

3.514 2.728 2.461 2.783 4.353 3.945 2.722 1.937 1.713 2.377 3.076 2.739 1.639 4.114

31.196 19.823 14.619 30.685 62.622 36.373 1.917 8.500 13.431 4.329 22.159 21.046 15.016 19.939

0.000 0.000 0.001 0.000 0.000 0.000 0.383 0.014 0.001 0.115 0.000 0.000 0.001 0.000

The results of the analysis for Group 1 and Group 2 can be provided upon request.

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

127

Table A.3 Nominal exchange rates — correlations.

Brazil Chile Colombia Czech. Rep. India Indonesia Rep. of Korea Mexico Philippines Poland Romania South Africa Thailand Turkey

Brazil

Chile

Colombia

Czech. Rep.

India

Indonesia

Rep. Of Korea

Mexico

Philippines

Poland

Romania

South Africa

Thailand

Turkey

1.00 0.84 0.87 0.38 0.10 0.02 0.28 −0.25 0.75 0.51 0.45 0.18 0.65 0.17

1.00 0.84 0.59 −0.02 0.05 0.55 −0.42 0.72 0.64 0.28 0.30 0.79 −0.03

1.00 0.48 −0.24 −0.16 0.28 −0.41 0.87 0.55 0.19 −0.09 0.78 −0.16

1.00 −0.14 −0.14 0.42 −0.76 0.39 0.91 −0.19 0.07 0.84 −0.58

1.00 0.62 0.20 0.53 −0.30 0.06 0.60 0.71 −0.30 0.71

1.00 0.21 0.38 −0.07 −0.15 0.29 0.63 −0.10 0.50

1.00 0.02 0.18 0.55 0.39 0.44 0.34 0.01

1.00 −0.40 −0.50 0.49 0.18 −0.77 0.71

1.00 0.41 0.19 −0.21 0.78 −0.15

1.00 0.13 0.15 0.74 −0.34

1.00 0.42 −0.09 0.76

1.00 −0.04 0.52

1.00 −0.48

1.00

Table A.4 Variation explained by the Exchange Rate Index. Brazil Chile Colombia Czech Republic India Indonesia Rep. of Korea Mexico Philippines Poland Romania South Africa Thailand Turkey

0.7256 0.8603 0.8679 0.8446 0.2134 0.1174 0.3222 0.7286 0.7935 0.8086 0.0009 0.0086 0.9717 0.3924

Table A.5 Granger causality I. VAR Granger causality/block exogeneity Wald tests Sample: 2000:01 2014:08 Included observations: 174 Dependent variable: Exchange Rate Index Excluded

Chi-sq

df

Prob.

EMBI+ SS Main VIX Risk premium EMBI+ SS main VIX Risk premium

1.562538 5.190001 2.720554 1.426989 5.232754 4.001306

1 1 1 2 2 2

0.2113 0.0227 0.0991 0.4899 0.0731 0.1352

Table A.6 Granger causality test II. VAR Granger causality/block exogeneity Wald tests (Lag = 2) Sample: 2000:01 2014:08 Included observations: 171 Dependent variable: Exchange Rate Index Excluded

Chi-sq

df

Prob.

Price level M1 Interest rate Inflation Industrial production All

5.335502 7.67991 0.538667 9.084272 4.134156 20.28819

2 2 2 2 2 10

0.0694 0.0215 0.7639 0.0107 0.1266 0.0266

128

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

Table A.7 Dynamic correlations of the Exchange Rate Index and financial variables.

VIX S&P500 EMBI+ SS main EMBI+ SS Latin EMBI+ SS Europe EMBI+ SS Asia MSCI emerging markets MSCI emerging Asia MSCI emerging Europe MSCI Latin

Long term correlation

Medium term correlation

Short term correlation

Static correlation

0.36 −0.68 0.58 0.65 0.39 0.55 −0.91 −0.88 −0.90 −0.86

0.57 −0.74 0.71 0.71 0.65 0.64 −0.89 −0.84 −0.91 −0.87

0.46 −0.51 0.62 0.59 0.61 0.54 −0.76 −0.70 −0.71 −0.71

0.49 −0.61 0.66 0.63 0.63 0.58 −0.82 −0.77 −0.80 −0.78

Table A.8 Uncovered equity return parity condition.

Coefficient Standard error Adjusted R2 Direction of change

Constant

Excess equity return

0.08 0.15 0.36 32%

−0.24 0.03

Table A.9 Linear and threshold regression results for the Exchange Rate Index and financial variables. Sample splitting⁎

Constant

VIX

R2

No. of obs.

Global OLS estimation

0.023 (0.159)

7.507 (1.815)

0.240

175

−0.110 (0.115) 0.198 (0.651)

2.263 (0.815) 18.432 (2.802)

0.487 0.045

154

0.697

21

Sample splitting

Constant

EMBI+ SS main

R2

No. of obs.

Global OLS estimation

0.101 (0.139)

17.123 (2.444)

0.435

175

−0.086 (0.106) 0.063 (0.391)

6.291 (1.569) 29.447 (2.314)

0.630 0.127

143

0.786

32

Threshold estimation for VIX Regime 1 — THRESHOLD ≤ 3.010 Regime 2 — THRESHOLD N 3.010

Threshold estimation for EMBI+ Regime 1 — THRESHOLD ≤ 2.547 Regime 2 — THRESHOLD N 2.547

⁎ This is a test of null of no threshold against alternative of threshold allowing for heteroskedastic errors (White corrected). We have 10,000 number of bootstrap replications and the trimming percentage is 0.15. LM test for no threshold for the VIX variable results with a bootstrap p-value of 0.00370. LM test for no threshold for the EMBI+ Sovereign Spreads variable results with a bootstrap p-value of 0.00000. Numbers in parenthesis are the standard errors and bold numbers indicate significant values. Table A.10 Markov switching results for the Exchange Rate Index and financial variables. Markov switching⁎

Constant

VIX

Sigma

R2

Regime 1

3.844 (0.542) −0.297 (0.144) 0.760 (0.134) 0.981 (0.014)

13.967 (1.725) 5.707 (0.945)

2.281 (1.008) 2.876 (0.337)

0.766

Markov switching⁎

Constant

EMBI+ SS Main

Sigma

R2

Regime 1

−0.544 (0.206) 0.682 (0.251) 0.916 (0.058) 0.925 (0.054)

6.927 (2.759) 24.028 (1.977)

2.077 (0.393) 2.510 (0.498)

0.643

Regime 2 p00 p11

Regime 2 p00 p11

0.628

0.432

⁎ Value of the log likelihood function for the model using EMBI+ Sovereign Spreads is −335.591, and the value of the log likelihood function for the model using VIX is −354.860. Numbers in parenthesis are the standard errors and bold numbers indicate significant values.

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

129

Table A.11 Linear and threshold regression results for the Exchange Rate Index and macroeconomic fundamentals. Sample splitting for PPP⁎

Constant

Prices

R2

No. of obs.

Global OLS estimation

−0.609 (0.215)

2.130 (0.432)

0.081

173

−0.746 (0.178) 5.260 (2.880)

1.840 −0.404 0.538 (1.640)

0.279 0.093

166

0.003

7

Threshold estimation for PPP Regime 1 — THRESHOLD ≤ 5.293 Regime 2 — THRESHOLD N 5.293

Sample splitting for SPM

Constant

Inflation

Output

M1

Interest rate

R2

No. of obs.

Global OLS estimation

0.064 (0.191)

1.420 (0.439)

−0.144 (0.346)

0.163 (0.148)

0.967 (0.557)

0.103

173

−0.197 (0.141) 6.620 (1.530)

0.771 (0.257) −2.100 (2.770)

−0.019 (0.127) −0.499 (0.645)

0.187 (0.094) 1.880 (0.730)

0.293 (0.476) 23.600 (3.730)

0.457 0.053

165

0.879

8

Threshold estimation for SPM Regime 1 — THRESHOLD ≤ 4.762 Regime 2 — THRESHOLD N 4.762

⁎ This is a test of null of no threshold against alternative of threshold allowing for heteroskedastic errors (White corrected). We have 10,000 number of bootstrap replications and the trimming percentage is 0.15. LM test for no threshold for the PPP model results with a bootstrap p-value of 0.00000. LM test for no threshold for the SPM model results with a bootstrap p-value of 0.00010. Numbers in parenthesis are the standard errors and bold numbers indicate significant values.

Appendix B. Figures

Fig. B.1. Standardized exchange rate series of emerging market countries.

130

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

Fig. B.2. Standardized exchange rate series of emerging market countries and the Exchange Rate Index.

Fig. B.3. Exchange Rate Index and major economic developments.

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

Fig. B.4. Exchange Rate Index and emerging market currencies.

131

132

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

Fig. B.5. Exchange Rate Index and emerging market currencies.

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

Fig. B.6. Exchange Rate Index and MSCI emerging markets.

Fig. B.7. Exchange Rate Index and EMBI+ sovereign spreads main.

Fig. B.8. Exchange Rate Index and VIX.

133

134

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

Dependent variable = Δ−f Independent variable = Δ−EMBI

Dependent variable = Δ−f Independent variable = Δ−VIX

16 LR (γ)

24

LR (γ)

n

n

15

95% critical value

95% critical value

22 14 20

13 12

Fn(γ)

Fn(γ)

18 16

11

14

10

12

9

10

8

8

7

6

6 1

2

3

4

5

γ

6

1

7

2

3

4

5

γ

6

7

Fig. B.9. Nonlinear test results for financial variables. Dependent variable = Δ−f Independent variable = PPP

Dependent variable = Δ−f Independent variable = SPM

10

15

9 8 7

10

Fn(γ)

Fn(γ)

6 5 4 5

3 2

LRn(γ)

LR (γ)

1

n

95% critical value

95% critical value

0 1

2

3

4

γ

5

6

7

8

0

1

2

3

4

Fig. B.10. Nonlinear test results for macroeconomic fundamentals.

Fig. B.11. Impulse response for PPP model.

γ

5

6

7

8

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

135

Fig. B.12. Impulse response for SPM model.

References Adrian, T., Etula, E., Shin, H., 2009. Risk appetite and exchange rates. Federal Reserve Bank of New York Staff Report, p. 361. Aggarwal, R., Simmons, W., 2008. Common stochastic trends among Caribbean currencies: evidence from Guyana, Jamaica, and Trinidad and Tobago. J. Econ. Bus. 60 (3), 277–289. Ang, A., Timmermann, A., 2012. Regime changes and financial markets. Annu. Rev. Financ. Econ. 4 (1), 313–337. Bai, J., Ng, S., 2007. Determining the number of primitive shocks in factor models. J. Bus. Econ. Stat. 25 (5), 52–60. Banbura, M., Giannone, D., Reichlin, L., 2010. Nowcasting. CEPR Discussion Papers DP7883. Bernanke, B.S., Boivin, J., Eliasz, P., 2005. Measuring the effects of monetary policy: a factor-augmented vector autoregressive approach. Q. J. Econ. 120 (1), 387–422. Bubula, A., Ötker, I., 2002. The evolution of exchange rate regimes since 1990: evidence from de facto policies. vol. 2. International Monetary Fund. Cairns, J., Ho, C., Mccauley, R., 2007. Exchange rates and global volatility: implications for Asia-Pacific currencies. BIS Quarterly Review (March). Cappiello, L., Santis, R.D., 2005. Explaining exchange rate dynamics: the uncovered equity return parity condition. ECB Working Paper 529. Cayen, J., Coletti, D., Lalonde, R., Maier, P., 2010. What drives exchange rates? New evidence from a panel of US dollar bilateral exchange rates. Bank of Canada Working Paper 5 (2). Chadwick, M., 2010. Performance of Bayesian dynamic latent factor models in measuring pricing errors and forecasting returns. (In PhD Thesis). University of London, Birkbeck College. Chen, S.L., Lee, H.Y., 2006. Why use Markov switching models in exchange rate prediction? Econ. Model. 23 (4), 662–668. Cheung, Y., Chinn, M., Pascual, A., 2005. Empirical exchange rate models of the nineties: Are any fit to survive? J. Int. Money Financ. 24 (7), 1150–1175. Chinn, M., Alquist, R., 2006. Conventional and unconventional approaches to exchange rate modeling and assessment. NBER Working Paper 12481.

Croux, C., Forni, M., Reichlin, L., 2001. A measure of comovement for economic variables: theory and empirics. Rev. Econ. Stat. 83 (2), 232–241. Dornbusch, R., 1976. Expectations and exchange rate dynamics. J. Polit. Econ. 84 (6), 1161–1176. Dornbusch, R., Fischer, S., 1980. Exchange rates and the current account. Am. Econ. Rev. 70 (5), 960–971. Doz, C., Giannone, D., Reichlin, L., 2006. A quasi maximum likelihood approach for large approximate dynamic factor models. CEPR Discussion Paper Series DP5724. Doz, C., Giannone, D., Reichlin, L., 2011. A two-step estimator for large approximate dynamic factor models based on Kalman filtering. J. Econ. 164 (1), 188–205. Dumas, B., Solnik, B., 1995. The world price of foreign exchange rate risk. J. Financ. 50 (2), 445–479. Engel, C., Mark, N., West, K., 2009. Factor model forecasts of exchange rates. In mimeo. University of Wisconsin. Fama, E., 1984. Forward and spot exchange rates. J. Monet. Econ. 14 (3), 319–338. Fiess, N., Shankar, R., 2009. Determinants of exchange rate regime switching. J. Int. Money Financ. 28 (1), 68–98. Frankel, J., 1979. On the mark: a theory of floating exchange rates based on real interest differentials. Am. Econ. Rev. 69 (4), 610–622. Franses, P., van Dijk, D., 2000. Non-linear Time Series Models in Empirical Finance. Cambridge University Press, Cambridge and New York. Frömmel, M., MacDonald, R., Menkhoff, L., 2005. Markov switching regimes in a monetary exchange rate model. Econ. Model. 22 (3), 485–502. Giannone, D., Reichlin, L., Small, D., 2008. Nowcasting: the real-time informational content of macroeconomic data. J. Monet. Econ. 55 (4), 665–676. Hansen, L., 1982. Large sample properties of generalized method of moments estimators. Econometrica 50 (4), 1029–1054. Hansen, B., 2000. Sample splitting and threshold estimation. Econometrica 68 (3), 575–603. Hodrick, R., 1989. Risk, uncertainty, and exchange rates. J. Monet. Econ. 23 (3), 433–459. Kal, S.H., 2011. Global capital flows, time-varying fundamentals and transitional exchange rate dynamics. J. Forecast. http://dx.doi.org/10.1002/for.1267.

136

M.G. Chadwick et al. / Economic Modelling 49 (2015) 120–136

Kempa, B., Wilde, W., 2011. Sources of exchange rate fluctuations with Taylor rule fundamentals. Econ. Model. 26 (6), 2622–2627. Kim, C.J., Nelson, C.R., 1998. Business cycle turning points, a new coincident index, and tests of duration dependence based on a dynamic factor model with regime switching. Rev. Econ. Stat. 80 (2), 188–201. Kim, C.J., Nelson, C.R., 2003. State-space models with regime switching: classical and Gibbs-sampling approaches with applications. MIT Press Books. Kose, A., Otrok, C., Whiteman, C., 2003. International business cycles: world, region and country-specific factors. Am. Econ. Rev. 93 (4), 1216–1239. Kose, A., Otrok, C., Whiteman, C.H., 2008. Understanding the evolution of world business cycles. J. Int. Econ. 75 (1), 110–130. Meese, R., Rogoff, K., 1983. Empirical exchange rate models of the seventies: do they fit out of sample? J. Int. Econ. 14 (1), 3–24. Meese, R., Rose, A., 1991. An empirical assessment of non-linearities in models of exchange rate determination. Rev. Econ. Stud. 58 (3), 603–619. Neely, C., Rapach, D.E., 2011. International comovements in inflation rates and country characteristics. Federal Reserve Bank of St. Louis, Working Paper 25. Obstfeld, M., Rogoff, K., 2000. The six major puzzles in international macroeconomics: is there a common cause? NBER Macroecon. Annu. 15, 339–390. Pan, M., Fok, R., Liu, Y., 2007. Dynamic linkages between exchange rates and stock prices: evidence from East Asian markets. Int. Rev. Econ. Financ. 16 (4), 503–520.

Rapach, D., Wohar, M., 2002. Testing the monetary model of exchange rate determination: new evidence from a century of data. J. Int. Econ. 58 (2), 359–385. Smith, R., Zoega, G., 2007. Global factors, unemployment adjustment and the natural rate. Kiel Working Paper Collection 2/1367. Stock, J., Watson, M., 1989. New indices of coincident and leading economic indicators. NBER Macroecon. Annu. 4, 351–394. Taylor, M., Peel, D., 2000. Nonlinear adjustment, long run equilibrium and exchange rate fundamentals. J. Int. Money Financ. 19 (1), 33–53. Tekatli, N., 2010. A Bayesian generalized factor model with comparative analysis. TCMB Working Paper Series 10 (18). Walid, C., Chaker, A., Masood, O., Fry, J., 2011. Stock market volatility and exchange rates in emerging countries: a Markov-state switching approach. Emerg. Mark. Rev. 12 (3), 272–292. Wang, Y., Wu, C., 2012. Energy prices and exchange rates of the us dollar: further evidence from linear and nonlinear causality analysis. Econ. Model. 29 (6), 2289–2297. Yuan, C., 2011. The exchange rate and macroeconomic determinants: time-varying transitional dynamics. North Am. J. Econ. Finance 22 (2), 197–220.