- Email: [email protected]
Real exchange rates and transition economies
Journal of International Money and Finance 56 (2015) 23e35
Contents lists available at ScienceDirect
Journal of International Money and Finance journal homepage: www.elsevier.com/locate/jimf
Real exchange rates and transition economies Gianna Boero a, Kostas Mavromatis b, Mark P. Taylor c, * a
Department of Economics, University of Warwick, UK Department of Economics and Econometrics, University of Amsterdam, The Netherlands c Warwick Business School, University of Warwick and Centre for Economic Policy Research, UK b
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 14 April 2015
In a number of empirical studies, transition economies have been shown to be subject to the Harrod-Balassa-Samuelson effect. This implies that the currencies of these countries have experienced a prolonged appreciation in real terms as their convergence proceeded. In this paper we find that a long-run relationship exists between the real exchange rate, productivity differentials, real interest rate differentials and the capital account for eight transition economies of Central and Eastern Europe, using monthly data over a period which extends from 1996 to 2013. We find that there are two sources of appreciation of the currencies of these countries, namely the Harrod-Balassa-Samuelson effect and the capital account effect, and argue that their significance depends on the type of investment received by the countries. While long-run foreign direct investment enhances productivity, porfolio investment leaves productivity unaffected, so our argument is that the larger foreign direct investment relative to portfolio investment, the greater the contribution of productivity in the determination of the real exchange rate. Moreover, we find that while the variables are linked by a linear long-run equilibrium relationship, adjustment towards equilibrium is nonlinear and is well represented by a smooth transition mechanism where the degree of equilibrium correction is a function of the sign and/or the size of the deviation from equilibrium. Interestingly, we find that a logistic smooth transition model fits well a larger number of countries, by allowing a different response of the real exchange rate to misalignments of different sign. © 2015 Elsevier Ltd. All rights reserved.
JEL classification: E52 F42 Keywords: Harrod-Balassa-Samuelson effect Real exchange rate Capital account Nonlinear models
* Corresponding author. Warwick Business School, University of Warwick, Coventry CV4 7AL, UK. Tel.: þ44 2476 524 534. E-mail address: [email protected] (M.P. Taylor). http://dx.doi.org/10.1016/j.jimonfin.2015.04.002 0261-5606/© 2015 Elsevier Ltd. All rights reserved.
24
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
1. Introduction The Harrod-Balassa-Samuelson effect, introduced by Harrod (1933), Balassa (1964) and Samuelson (1964), states that countries that have high productivity gains experience prolonged real appreciations of their currencies against the currencies of less productive countries. A vast empirical literature focusing either on transition or on industrialized economies supports this perspective. For example, Halpern and Wyplosz (1996), using panel data techniques for a number of transition economies find significant evidence in favour of the Harrod-Balassa-Samuelson effect at a relatively early stage of the transition process. More recently, though, there has been a growing recognition that the HarrodBalassa-Samuelson effect accounts only partly for the observed appreciation of the currencies of transition economies. A detailed overview of this literature, covering both theoretical and empirical aspects of recent contributions, including several additional explanations for the movements of real exchange rates in transition economies, can be found in Egert et al. (2006a). Two recent studies by Babecky et al. (2009, 2010) offer interesting insights on the interaction between foreign direct investment, net exports and sustainable real exchange rates in the new EU member states, and their findings are particularly relevant for the focus of our paper. Using a calibrated and simulated theoretical model, these studies conduct a sectoral analysis of foreign direct investment (FDI) and argue that transition economies where FDI is absorbed by the tradable sector are likely to benefit substantially from capital inflows and experience real currency appreciation. Building upon this literature, using time series econometric techniques, in this paper we take an innovative approach to investigate the way the composition of the capital account affects productivity and the real exchange rate in transition economies. We explore this channel by revisiting the behaviour of the real exchange rate in eight Central and Eastern European (CEE) countries, namely the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, the Slovak Republic and Slovenia, using monthly data over a sample period which goes from 1996 to the end of 2013. This period covers relatively early stages of the transition from planned to market-based economies, includes the entire period of preparation for accession to the European Union, which for these eight countries took place on 1 May 2004, and extends to the most recent years covering the Great Recession of 2008e09, which has clearly had an impact on the state of the convergence of these countries. We begin our empirical analysis by examining whether a long-run relationship exists among the real exchange rate, the productivity differential, the real interest rate differential and the capital account, and find supportive evidence of large productivity differential and capital account effects on the real exchange rate for a number of these transition economies. In addition, we show that the capital account effects depend on the composition of the country's capital inflows. In particular, we find that the capital account plays a role in the appreciation of the currency only to the extent that foreign direct investment exceeds portfolio investment. This is because FDI enhances productivity, while productivity is not affected by portfolio investment and, consequently, economies in transition can achieve a faster convergence by attracting long-run productivity-enhancing FDI. Long-term capital flows are the FDI which fosters growth, as it is generally taken to be determined by long-term profitability considerations and often leads to the transfer of state-of-the-art technology. As a result, these kinds of flows are less subject to market sentiment. On the contrary, portfolio investment and short-term bank lending (financial products) are short-run investments and subject to any kind of market volatility. Therefore, if portfolio investment is the key component of capital inflows to a transition economy, then it is likely that productivity is less strengthened compared to developed countries. Secondly, having established a long-run relationship between the real exchange rate and the variables mentioned above, namely, the productivity differential, the real interest rate differential and the capital account, we explore the possibility that the variables cointegrate in a nonlinear fashion, such that the speed of adjustment towards the long-run relationship varies over time, depending on the size and/or the sign of the exchange rate misalignment. A good deal of evidence in favor of nonlinear dynamical structure for the exchange rate and other macroeconomic variables has been reported in the literature. For example, in an application to transition economies during the 1990s, Taylor and Sarno (2001) find that the real exchange rate and the real interest rate differential cointegrate in a nonlinear fashion. Lothian and Taylor (2008) also employ models with nonlinear adjustment to test for
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
25
the effects of productivity differentials on the real exchange rate in developed economies and find cases where the variables cointegrate in one regime but not in the other. In line with the findings of these studies, our statistical analysis, based on a longer and more recent sample period for transition economies, provides strong evidence of nonlinear adjustments of the exchange rates in all countries under consideration. The remainder of the paper is organized as follows. In Section 2 we describe the environment in transition economies and set out the long-run equilibrium relationship which forms the basis of our subsequent analysis. In Section 3 we present the dataset and in Section 4 the econometric modelling strategy. In Section 5 we discuss the empirical results and Section 6 concludes. 2. The environment in transition economies After the initial recession, transition economies have been characterized by substantial productivity gains particularly in their industrial sector, coupled with a steady increase in the relative price of nontradables and a trend appreciation in the real exchange rate. So, the Harrod-Balassa-Samuelson (HBS) effect can be expected to be at work in these economies, and indeed, their currencies have been the subject of a large number of studies estimating the extent to which the HBS effect is reflected in real exchange rate movements. More precisely, in the literature, researchers assume that there are two sectors, tradables and non-tradables and that the increase in productivity takes place in the tradables sector, which consists of goods that are exported. A rise in productivity causes a rise in wages in the tradables sector, while prices remain unchanged due to perfect competition in the market for goods that are traded internationally. At the same time, the labour market is assumed to be perfectly competitive and labour is perfectly mobile within each country, but not across countries. Consequently, wages in the tradables and the non-tradables sectors are equalised. This will cause an increase in the price of non-tradables with a consequent rise in the overall price index. Assuming Purchasing Power Parity (PPP) holds in the tradables sector, the overall rise in the price level e at the prevailing nominal exchange rate e will result in an appreciation of the domestic currency in real terms. Before the start of the convergence process, transition economies experienced a depreciation of their currencies, sometimes quite abruptly, which was then followed by a systematic appreciation. One of the reasons for the appreciation of the currency was the increased labour mobility observed towards the more productive sectors, as inefficient production units shut down. Moreover, this appreciation of the real exchange rate has been further supported by inflows of direct investment. As productivity levels were converging to those of the industrialized countries, the rates of appreciation of the currencies started to slow down. This gradual phasing out of the appreciation of the real exchange rate experienced by the transition economies under consideration is illustrated in the plots in Fig. 1. For most countries, the real exchange rate appreciated steadily from the beginning of the sample to 2002, and then flattened considerably afterwards or even reversed its course, around the years of the Great Recession, turning into gradual depreciation in some of the countries. The real exchange rate is constructed, in logarithmic form, as follows:
q ¼ e þ p* p
(1)
where e is the logarithm of the nominal exchange rate, defined as the domestic price of one unit of foreign (Euro) currency, and p* and p are the logarithms of the foreign (Eurozone) and domestic price levels, respectively. Thus, an increase in the real exchange rate q corresponds to a depreciation, while a fall indicates an appreciation. Since, the purpose of the present paper is to examine whether the dynamics of the real exchange rate of transition economies are determined by factors other than the HBS effect, as the starting point of our analysis we consider the following long-run equation:
qt ¼ a0 þ a1 yt þ a2 rt rt* þ a3 CAt þ ut
(2)
where ut is an error term, yt is the logarithm of relative productivity (domestic productivity relative to foreign productivity), rt rt* is the real interest rate differential, where rt and rt* denote the domestic and foreign real interest rate respectively, and CAt denotes the capital account to GDP ratio. This relationship is very close to a standard specification based on the stock-flow approach to the real
26
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
Fig. 1. (Log) Real exchange rates.
exchange rate, advocated, for instance, by Alberola et al. (1999). According to the HBS effect, and other explanations focusing on the tradable-based real exchange rate, the coefficient on relative productivity is expected to be negative, that is, a rise in relative productivity causes a real exchange rate appreciation induced through relative price adjustment. On the other hand, New Open Economy Macroeconomics models show that productivity gains can translate into a real depreciation of the tradable-deflated real exchange rate through the terms-of-trade channel, and the overall impact on the real exchange rate of the whole economy will depend on which channel prevails (see Egert et al., 2006a; Egert et al., 2006b). In this paper, we argue that the composition of capital flows into the transition economies plays an important role in the explanation of the exchange rate dynamics of these economies. Below, we show that both the coefficient on the capital account and on relative productivity (HBS effect) depend on the type of investment which flows into these economies. During their transition process, these economies received large inflows of investment and, consequently, their capital accounts experienced long-lasting surpluses. However, not all capital inflows enhance productivity. The latter is increased with FDI, whereas portfolio investment (PI) does not contribute to productivity, and therefore capital account surpluses do not necessarily translate into higher productivity. Finally, as capital market integration is still an ongoing process in these economies, interest rate differentials may be an additional factor underlying the behavior of the real exchange rate, and is therefore included in our model. According to the traditional Mundell-Fleming-Dornbush model, real exchange rates and real interest rate differentials should be inversely related and move in opposite directions in the long run (see, e.g., Sarno and Taylor, 2002, ch 4).
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
27
3. The data We consider the following countries: the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, the Slovak Republic and Slovenia, and compare the first six countries with the Eurozone-17, while the last two are compared with the Eurozone-12.1 The data are taken at a monthly frequency from the International Monetary Fund International Financial Statistics, the Organization for Economic Cooperation and Development Main Economic Indicators and Eurostat. We use the end-of-period nominal exchange rate of the currency of each country against the Euro. For the period before the functioning of the Eurozone we use the ECU. The long-term government bond rate is used as a proxy for the nominal interest rate. The German 10-year government bond rate is used as a proxy for the long-term rate in the Eurozone. The consumer price index (CPI) is used as a proxy for the price level for each of the transition economies, while the harmonized index of consumer prices (HICP) is used as a proxy for the price level in the Eurozone. We measure productivity as the ratio of industrial production, excluding construction, to the number of people employed in industry. The capital account is measured as domestic assets held by foreign residents minus foreign assets held by domestic residents expressed in terms of GDP.2 Our sample starts from 1996 in order to eliminate the early years of transition during which productivity improvements were the result of initial reforms rather than the HBS effect, and spans through 2013.3 Both the start and end dates for each country is based on data availability. In particular, the sample periods used are the following: for the Czech Republic 1996:1e2013:2, for Estonia 1997:4e2011:7, for Hungary 1996:1e2013:7, for Latvia 1996:2e2013:9, for Lithuania 1997:1e2011:7, for Poland 1996:1e2013:9, for Slovenia and Slovak Republic 1996:1e2012:3. 4. Econometric modelling strategy The first step in our empirical strategy is to test whether Equation (2) constitutes a long-run relationship among the four variables considered. For this purpose we apply the DickeyeFuller and the Johansen test for cointegration. A finding of cointegration implies that the variables have an error correction representation, allowing for the long-run equilibrium relationship and at the same time capturing the short-run dynamics of the variables. A number of studies have reported evidence in favor of non-linear adjustments of the exchange rate towards its equilibrium value (see, amongst others, Taylor and Sarno, 2001, for transition economies, and Lothian and Taylor, 2008, for developed countries). If we find evidence of cointegration, then we explore the possibility that the mean reversion properties of the cointegrating residuals vary with the extent of the deviations from equilibrium. A suitable framework for modelling potential nonlinearities of this kind is the Smooth Transition Autoregressive (STAR) model. These models have received increasing attention in the econometric time series literature with applications to exchange rates and other macroeconomic variables.4 Our €svirta empirical analysis is conducted within this setting and follows the procedure suggested by Tera (1994). The procedure starts with the detection of nonlinearities in the cointegrating residuals of Equation (2), using the tests described in Luukkonen et al. (1988). The null hypothesis of linearity is tested against the alternative of an exponential smooth transition autoregressive (ESTAR) model or a logistic smooth transition autoregressive (LSTAR) model. These models describe, in different ways, a smooth transition between two extreme regimes for the adjustment of the real exchange rate. In the ESTAR model, the adjustment is symmetric, in the sense that only the distance from equilibrium matters, but not the sign. For small deviations from equilibrium, the model depicts the 'inner regime', while at the other extreme, for large misalignments above and below the equilibrium level, the model
1 Following the suggestion of an anonymous referee, we compare the Slovak Republic and Slovenia to the Eurozone (12 members) since Slovenia was the 13th member while Slovak Republic was the 16th. 2 Monthly GDP was obtained by linear interpolation from the quarterly series. 3 In an earlier version of the paper we included the years before 1996. However, following the suggestion of an anonymous referee, we have now excluded those initial years from our sample. 4 Taylor and Sarno (2001) found that the adjustment process of the real exchange rate in nine transition economies over the period leading to the end of 1997 was well described by an exponential smooth transition autoregressive process (ESTAR).
28
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
depicts the 'outer regime'. As the exchange rate moves further away from equilibrium, the model describes a nonlinear adjustment process with mean-reverting behavior which varies with the size of the deviation from equilibrium. The LSTAR model describes an asymmetric adjustment. The model parameters smoothly change between two extreme regimes, one characterized by large appreciations, the other by large depreciations, with one regime having more impact than the other according to the values of the misalignment. Formally, the STAR model for the cointegrating residuals ut is represented as follows:
u t ¼ c0 þ
Xp
cu i¼1 i ti
þ ðc00 þ
Xp
c0 u ÞGðutd ; k; cÞ þ et i¼1 i ti
(3)
This model consists of two autoregressive parts linked by a continuous nonlinear transition function, G(utd;k,c), bounded between zero and one, allowing for changes in the model parameters according to the value of the transition variable utd relative to a threshold value c. The transition function is exponential G(utd;k,c) ¼ 1 exp(k(utd c)2) in the ESTAR model, and logistic G(utd;k,c) ¼ (1 þ exp(k(utd c)))1 in the LSTAR model. These functions describe, in different ways, a smooth transition of ut between two extreme regimes associated with the extreme values of the transition functions. The parameter k > 0 is a transition parameters that governs the speed of transition between regimes, with high values implying a fast adjustment between regimes; et is an iid process; p is the lag-length, and d is a delay parameter. In the ESTAR model, for small deviations from equilibrium, as utd approaches c, the transition function approaches zero, and the model depicts the 'inner regime' where the series is described by a linear autoregressive model with coefficients c0 and ci, i ¼ 1,...,p. As utd moves further away, above or below the threshold value c, the model describes a nonlinear adjustment process with a mean-reverting behavior which varies with the size of the deviation from equilibrium. At the extreme, in the 'outer regime', for very large misalignments of both signs, the transition function approaches one, and the series is described by an autoregressive model with 0 0 intercept equal to c0 þ c0 and autoregressive coefficients equal to ci þ ci for i ¼ 1,...,p. Also, the model depicts a linear autoregression as k approaches zero or infinity, as the transition function becomes constant (equal to zero and one respectively). The LSTAR model describes an asymmetric adjustment toward c. The logistic transition function changes monotonically from zero to one as utd increases, and G(utd;k,c) ¼ 0.5 when utd is equal to c, the threshold between the two regimes. As in the ESTAR model, in the limit, as k approaches zero or infinity, the value of the transition function is constant, so the model becomes a linear autoregressive process. For intermediate values of k, the transition function approaches zero for values of utd far below c (large appreciations), and approaches one for values far above c (large depreciations). The model parameters smoothly change between these two extremes, with one regime having more impact than the other according to the values of utd c. The following reparameterization of model (3) is often used when studying the mean reversion properties of the series, when the true data generating process is an ESTAR or an LSTAR process:
Dut ¼ w0 þ wut1 þ
X p1 i¼1
wi Duti þ ðw00 þ w0 ut1 þ
X p1 i¼1
0
w0i Duti ÞGðutd ; k; cÞ þ et 0
(4)
0
This STAR formulation where w ¼ (c1 þ ... þ cp 1) and w ¼ ðc1 þ ::: þ cp 1Þ will display global 0 stability as long as ðw þ w Þ < 0. However, this condition does not rule out the possibility that the process is locally nonstationary, with w ¼ 0, implying unit root behavior in the linear part, or even the possibility of locally explosive dynamics, with w > 0. Following Luukkonen et al. (1988), a test for the null of linearity against the alternative of either an ESTAR or LSTAR stationary model can be computed using the auxiliary regression:
ut ¼ b1 zt þ b2 zt utd þ b3 zt u2td þ b4 zt u3td þ xt
(5)
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
29
0
where zt ¼ ½1; ut1 ; …; utp , the auxiliary parameters bi ¼ [bi0, bi1,..., bip], for i ¼ 1, 2, 3, 4, are functions of the parameters of the original STAR model, and xt is an error term.5 The test for the null hypothesis of linearity against STAR nonlinearity consists of testing the following restrictions:
HL : b2 ¼ b3 ¼ b4 ¼ 0 The order p of the linear autoregressive model is selected on the basis of the partial autocorrelation function of the series, and the delay parameter d for the transition variable utd is selected according to the procedure suggested by Ter€ asvirta (1994) as the value that produces the strongest rejection (the smallest p-value) of the null hypothesis of linearity. If the null hypothesis is rejected, the next step is to decide whether an ESTAR or an LSTAR is the appropriate model. The proposed decision rule consists of testing the following sequence of null hypotheses:
H01 : b4 ¼ 0; H02 : b3 ¼ 0 j b4 ¼ 0; H03 : b2 ¼ 0 j b4 ¼ b3 ¼ 0 The above sequence of tests can be performed as conventional Lagrange multiplier (LM) tests with an asymptotic chi-square distribution or using the F version of the LM test, which is preferred in small samples. If the p-value of the test corresponding to H02 is the smallest among the three tests, the ESTAR model should be selected, while in all other cases an LSTAR model is the appropriate choice. If nonlinearity in the cointegrating residuals is detected, then a nonlinear version of the error correction model, where the real exchange rate adjusts nonlinearly to past deviations from its long-run equilibrium, would provide a suitable description of its short-run dynamics. 5. Empirical results 5.1. The long-run equation In this section we present the results of the estimation of the long-run relationship (Equation (2)) and analyze the mean reverting properties of the residuals of this equation. In order to test the validity of Equation (2) as a long-run equilibrium relation, we start by testing whether the real exchange rate (qt), relative productivity (yt), interest rate differential ðrt rt* Þ and capital account (CAt) are cointegrated. Stationarity tests (available upon request) suggest that all four variables are integrated of order one; hence, we proceed to test for the existence of a common trend among them. We use both the DickeyeFuller test on the residuals of the long run regression and the Johansen cointegration procedure. The latter performs reasonably well in terms of power and size, when the adjustment towards equilibrium is nonlinear (see Balke and Fomby, 1997). We computed both the maximum eigenvalue test and the trace test and obtained similar results, therefore, in Table 1 (Panel A) we report only the results from the former test. While the DickeyeFuller tests were unable to reject the null hypothesis of non-stationarity in most cases, the existence of one cointegrating vector could be established with the Johansen test for all countries either at the 1-percent or at the 5percent significance levels. The maximum likelihood estimates of the cointegrating coefficients, based on the VAR specification, are reported in Panel B of Table 1. Any potential endogeneity of the variables in the model is taken into account by the Johansen approach used to estimate the long-run coefficients. This means that we account for issues of potential simultaneity that could arise, for example, from productivity being determined by the capital account and vice-versa. Therefore, the VAR approach is particularly suitable given that productivity gains have been driven by large capital inflows in these countries, and these two variables are potentially strongly correlated. Essentially, by considering the CAt variable in the long-run equation together with the productivity variable, we control for the potential effects of capital flows on productivity, and therefore the coefficient of CAt can be interpreted as an estimate of its impact
5
The auxiliary regression is based on a third-order Taylor series approximation of the transition function.
30
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35 Table 1 Cointegration results. Panel A: Cointegration test statistics
Czech Republic Estonia Hungary Latvia Lithuania Poland Slovak Republic Slovenia
D-F test
Joh. test: max (r ¼ 0 vs r ¼ 1)
2.13710 2.87483 3.14199* 1.83711 1.40915 2.83779 2.35878 2.45255
35.0052*** 28.8818** 32.6793** 59.4450*** 31.7927** 40.3253*** 47.8784*** 29.2182**
Panel B: Estimated cointegrating vectors
Czech Republic Estonia Hungary Latvia Lithuania Poland Slovak Republic Slovenia
yt
(r r*)t
C At
Constant
0.849 1.668 0.255 1.231 0.259 0.195 0.400 0.569
0.051 0.060 e 0.105 0.017 0.189 0.036 0.070
0.860 0.034 0.083 0.341 0.040 0.199 0.010 0.024
3.489 0.598 e 0.408 0.135 0.943 e 0.292
Notes: Panel A:***Significant at 1%. **Significant at 5%. *Significant at 10%. For the Johansen test (Joh. Test), the critical values for the max test statistics, for H0: r ¼ 0 vs r ¼ 1, are 33.24 for the 1%, 28.14 for the 5% and 25.56 for the 10% significance level, respectively (Johansen and Juselius (1990)). Panel B: Maximum Likelihood estimates. The cointegrating vectors are derived from the Johansen procedure and normalized so that to receive the parameter values corresponding to Equation (2).
on the real exchange rate, once the indirect effect on productivity has been taken into account. The estimation results in Table 1 show a negative coefficient for productivity for all countries, implying that the currencies of these countries are subject to the HBS effect. As the catching-up process of those economies continues, increases in relative productivity induce an appreciation of the real exchange rate. This result is in line with the recent literature. For example, in a panel study for 11 transition economies, Egert et al. (2006b) provide evidence in favor of a negative effect of productivity on the real exchange rate.6 Table 1 also shows that the value of the coefficient on relative productivity differs substantially across countries, revealing that the HBS effect is more important for some countries (Czech Republic, Estonia, Latvia, Slovak Republic and Slovenia) than for others (Hungary, Lithuania and Poland). The real interest rate differential has a negative sign except for Poland, Slovak Republic and Slovenia, and the coefficient of the capital account variable is negative for all countries, except for Hungary Lithuania and Poland. As real exchange rate appreciation is the equilibrium behavior of the exchange rate of these economies over most of the period considered, the currencies of the countries where the coefficient on the capital account is negative will have two sources of appreciation: the productivity differential and the capital account surplus (or the current account deficit). An examination of how FDI and PI evolved through time in each country provides useful insights for the interpretation of the results obtained from the estimation of Equation (2). Our findings are in line with some of the recent literature which has investigated the direct link between FDI and productivity. For example, Babecky et al. (2009) conduct an analysis based on a structural medium-term model of sustainable real exchange rates and argue that inflows of foreign direct investment tend to appreciate the currencies of those
6 Their sample, though, spans until 2004, while their emphasis is on net foreign assets, rather than on the capital account as a whole. Moreover, they use the real effective exchange rate for each country instead of comparing the currencies of those economies to the Euro as is the case in this paper.
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
31
economies in real terms, as long as they are absorbed by the tradable goods sector.7 Our results suggest that for the countries with a negative coefficient on the capital account, the effect of productivity (the HBS effect) is accentuated with respect to those countries where the capital account surpluses cause real exchange rate depreciation. A further examination of the effects of the capital account conducted in the next section suggests that the coefficient on productivity may depend on the kind of investment that each country receives. In particular, we argue that for countries with high levels of FDI relative to portfolio investment, productivity proves to be the main driving force behind the prolonged real appreciation of their currencies, while this is not the case when FDI and portfolio investment are equally important. 5.2. What explains the long-run coefficient on the capital account and productivity? In order to determine the importance of the capital account in the determination of the real exchange rate in the transition economies, we examine the evolution of FDI and portfolio investment (PI) over the period considered. While FDI enhances productivity, PI leaves productivity unaffected. We collected data on net inward FDI and PI for each economy as a percentage of GDP. For countries in which FDI represents, on average, a substantially higher percentage of GDP than PI, the coefficient on the capital account in the long-run Equation (2) exhibits a negative value. This is the case for the Czech Republic, Latvia, the Slovak Republic and Slovenia, and these are also the countries with a higher negative coefficient on productivity differentials. This means that these countries have received important waves of labour augmenting productivity investment. For example, for the Czech Republic, FDI amounts to 4.57% of GDP and PI only to 0.40% on average over the sample period, and for the Slovak Republic, FDI is 3.28% of GDP and PI only 0.38%. On the other hand, for Poland, Hungary and Lithuania, FDI and PI are much closer, on average, and the effects of FDI are likely to be offset by those of PI. For Poland, for example, FDI amounts to 2.8% of GDP and PI to 2.28%, for Hungary, FDI amounts to 2.66% and PI to 1.76%, while for Lithuania, FDI amounts to 3.58% and PI to 4.94%. For these three countries, the capital account coefficient in the long-run relationship is positive, in contrast to the other countries; moreover, the coefficient on the productivity differential is lower than in the other countries. Overall, this analysis supports our conjecture that high levels of FDI, relative to portfolio investment, strengthen the effect of productivity on the real exchange rate, alongside the usual HBS effect. On the other hand, when FDI and PI are equally important, then financial factors play a significant role in the determination of the real exchange rate, which may be reflected in a reduced long-run effect of productivity. Previous studies based on a decomposition of FDI have shown that while FDI in the tradables sector leads to an improvement of the external balance, creating economy wide externalities, when FDI inflows are mainly absorbed by the non-tradable goods and services sectors, then such inflows could be associated with a deterioration of the trade balance, which would require a real exchange rate depreciation in order for the economy to return to a sustainable path of economic development (see Kinoshita, 2011; Babecky et al., 2009, 2010; Aykut and Sayek, 2007). In such a case, relative productivity will have either a weak negative effect on the real exchange rate (i.e. appreciation) or a positive effect (i.e. depreciation). In order to explore this issue further, for each country we constructed indices of productivity for the manufacturing (tradables) and construction (non-tradables) sectors and compare their evolution over time. As shown in Fig. 2, for Hungary, Lithuania and Poland, productivity in the construction sector is higher relative to that in the manufacturing sector during most of the sample period considered, which may reflect the fact that FDI is absorbed to a higher proportion by the nontradables sector, and may be taken as an additional factor explaining the estimated positive capital account coefficient for these countries. Thus, when FDI inflows are mainly absorbed by the nontradable goods and services sectors, such inflows could be associated with a deterioration of the trade balance, which would require a real exchange rate depreciation in order for the economy to
7 On the other hand, Egert et al. (2006b) show that an increase in net foreign liabilities implies a real exchange rate depreciation for the countries considered in our paper with the exception of Latvia. However, their analysis is conducted in a different context as they consider an earlier period, which starts in 1973, and use the real effective exchange rate as the variable of interest.
32
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
Fig. 2. Productivity in manufacturing and construction industry.
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
33
Table 2 Linearity of the cointegrating residuals.
Czech Republic Estonia Hungary Latvia Lithuania Poland Slovak Republic Slovenia
HL (d)
H01
H02
H03
0.0175( b d ¼ 14) 0.0103 ( b d ¼ 8) 0.0000 ( b d ¼ 6) 0.0112 ( b d ¼ 3) 0.0212 ( b d ¼ 3) b 0.0254 ( d ¼ 3) 0.0059 ( b d ¼ 8) 0.0028 ( b d ¼ 6)
0.6909 0.2783 0.0000 0.5204 0.0068 0.0230 0.2820 0.4264
0.0072 0.0370 0.0261 0.0022 0.2378 0.0436 0.1039 0.0134
0.0968 0.0162 0.4891 0.2459 0.3265 0.8274 0.0032 0.0065
Notes: P-values of the LM test statistics described in Section 4. If the P-value of the test corresponding to H02 is the smallest the ESTAR model is selected, otherwise the LSTAR model is the appropriate choice.
return to a sustainable path of economic development. At the same time, if the economy receives equally high portfolio investment (as is the case with Hungary, Lithuania and Poland), it needs a trade surplus in order to be able to pay back its external debt.8 Unless FDI inflows go to tradables, the economy would need a real exchange rate depreciation in order to satisfy this requirement. 5.3. Nonlinear short-run dynamics In this section we present the results of an investigation of the short-run dynamics of the exchange rates for the eight transition economies considered. Having found a long-run equilibrium relationship, we explore the possibility that the short-run adjustment towards equilibrium varies with the extent of the deviations from equilibrium. In particular, we explore the existence of potential STAR nonlinearity in the cointegrating residuals of Equation (2), following the procedure described in Section 4. The results of the linearity tests for each country and the delay parameter of the transition variable, d, are presented in Table 2. The null hypothesis of linearity against a STAR alternative is rejected at the 5% or 1% level of significance for all countries, implying the existence of nonlinear behavior of the short-run adjustment process. The delay parameter in some cases is high (d ¼ 14 for the Czech Republic, and d ¼ 8 for Estonia and the Slovak Republic), reflecting a slow response to shocks in the real exchange rate of these countries. For most countries, the smallest p-value is found for either H01 or H03, and this result, €svirta decision rule, suggests an LSTAR specification. The only two countries for according to the Tera which the smallest p-value occurs for H02, suggesting an ESTAR specification, are the Czech Republic and Latvia. As explained in Section 4, LSTAR nonlinearity implies an asymmetric adjustment of the cointegrating residuals, depending on both the size and the sign of past disequilibria, while ESTAR nonlinearity implies a symmetric adjustment for deviations above and below the equilibrium level. We then fit ESTAR or LSTAR models to the cointegrating residuals of each country, according to the results of the linearity tests. The models are estimated using the reparameterization (4) described in Section 4. In Table 3 we report the estimates of the coefficients w and w', providing information on the degree of mean reversion across regimes, with p-values in brackets. All models show global stability, 0 that is ðw þ w Þ < 0. This is not surprising, as we would expect cointegrating residuals to be globally stationary. However, there are cases where the process is locally nonstationary, indicating that adjustment to the long-run equilibrium occurs in one regime, but not in the other. For example, the ESTAR model for the Czech Republic indicates non-mean reversion (lack of cointegration) in the inner regime, that is, in the neighborhood of the equilibrium, and mean reversion (the variables cointegrate) in the outer regime, characterized by large misalignments of both signs, a result which is consistent with previous studies. Also interestingly, the LSTAR models for Lithuania and for the Slovak Republic show non-mean reversion in the extreme regime characterized by large over-valuations, and mean reversion in the other extreme regime characterized by large under-valuations. The remaining results
8 Further research might usefully examine the composition of portfolio flows in terms of equity and debt, and whether this has a bearing on this analysis.
34
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
Table 3 Estimated STAR models for the cointegrating residuals. w Czech
e
Estonia
0.177 (0.010) 0.112 (0.441) 0.194 (0.031) 0.071 (0.167) 0.106 (0.031) 0.182 (0.113) 0.671 (0.000)
Hungary Latvia Lithuania Poland Slovak Republic Slovenia
w0
kE
0.184 (0.083) 0.163 (0.097) 0.194 (0.090) 0.193 (0.017) 0.500 (0.000) 0.638 (0.000) 0.524 (0.024) 0.462 (0.002)
0.800 [0.023]
kL
0.936 [0.010] 0.871 [0.015] 0.923 [0.034] 0.144 [0.023] 2.339 [0.035] 0.650 [0.003] 0.721 [0.019]
R2
LMARCH
TS
0.176
{0.367}
{0.175}
0.116
{0.411}
{0.207}
0.167
{0.064}
{0.159}
0.181
{0.504}
{0.081}
0.121
{0.133}
{0.802}
0.096
{0.849}
{0.466}
0.160
{0.459}
{0.113}
0.288
{0.529}
{0.904}
Notes: P-values in parentheses. kE and kL are the transition parameters for the ESTAR and LSTAR model respectively, and the numbers in square brackets are the empirical P-values obtained from Monte Carlo simulations. The last two columns report the P-values for the LM test for autoregressive conditional heteroskedasticity and the Tsay test (TS) for neglected nonlinearities (Tsay (1986)).
for the other countries show that the real exchange rate adjustment in all cases is highly nonlinear, and the type and speed of adjustment is country specific, reflecting different reactions to correct misalignments of different sign and size. Table 3 also reports the estimated transition parameters kE (for the ESTAR models) and kL (for the LSTAR models), implying a moderate speed of transition between regimes for most countries, the highest being in the LSTAR model for Poland (kL ¼ 2.34). The last two columns of Table 3 report diagnostic tests showing that the estimated models perform satisfactorily. Our finding of LSTAR nonlinearity for the majority of the transition economies considered in this study is new, and suggests a different dynamics from that found in previous studies, which have reported evidence of symmetric ESTAR nonlinearity (see, for example, Taylor and Sarno, 2001). LSTAR models allow the exchange rate to respond differently not only to the size of the deviations from equilibrium, but also to over-valuations and under-valuations. This type of asymmetry seems more plausible over a period of such significant changes for these economies which have been subject to pressures of various kinds at different stages of their transition process, to the challenges posed by accession to the EU, and more recently by the Great Recession. Our results are indicative of the complexity of the economies of these countries, and LSTAR nonlinearity may better reflect policy interventions which react differently to correct misalignments of different size and sign. A full exploration of the short-run dynamics, allowing also for the possibility of endogeneity among all of the four variables considered in this studydsuch as the extent to which productivity gains are driven by capital inflows, for exampled, would require the use of a multivariate smooth transition equilibrium correction model. However, this analysis goes beyond the scope of the present paper, and represents an avenue for future research. 6. Conclusion The behavior of the real exchange rate in transition economies, particularly during the transition and catching-up process, has been the focus of several studies. In this paper we have reexamined the behavior of the real exchange rate in eight transition economies of Central and Eastern Europe, using a data set which extends from the early period of sustained appreciation in the mid 1990s, to the most recent period, characterized by a gradual phase-out of the real appreciation. The empirical contribution of this paper is twofold. First, we showed that there are two sources of appreciation of the real exchange rate: the HBS effect, via productivity differentials, and capital account effects. In particular, we argued that the HBS effect depends itself on the composition of the capital flows in each country: high levels of FDI relative to portfolio investment strengthen the effect of productivity, and hence the HBS
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
35
effect on the real exchange rate is accentuated. On the other hand, the HBS effect is weakened in countries where foreign direct investment are on average, over the sample period, close to portfolio investment; in this case, financial factors are equally important in determining the real exchange rate, thereby reducing the net effect of productivity. We then showed that while the variables are linked by a linear long-run equilibrium relationship, the adjustment of the real exchange rate towards equilibrium is nonlinear and is well described by a smooth transition mechanism where the degree of equilibrium correction is a function of the sign and/ or the size of the deviation from equilibrium. Interestingly, we find that for most countries a logistic smooth transition model fits well the short run dynamics of the real exchange rate, allowing a different response to misalignments of different sign. Our finding of LSTAR nonlinearity for the majority of the transition economies considered in this study is new, and suggests a different dynamics from that found in previous studies, which have reported evidence of symmetric ESTAR nonlinearity. LSTAR models allow the exchange rate to respond differently not only to the size of the deviations from equilibrium, but also to over-valuations and under-valuations. This type of asymmetry may reflect policy interventions which react differently to misalignments of different sign, and seems plausible for economies which have been exposed to shocks of various kinds at different stages of their transition process, to the challenges posed by accession to the EU, and more recently by the Great Recession. Now, more than ten years after joining the EU (May 2004), these economies are still in the process of catching up to the EU level of development. Acknowledgments The authors are grateful for constructive comments on an earlier version of this paper from the Editor, Menzie Chinn, and from an anonymous referee; any errors that remain are the responsibility of the authors alone. References Alberola, E., Cervero, S.G., Lopez, J.H., Ubide, A.J., Dec. 1999. Global Equilibrium Exchange Rates e Euro, Dollar, “Ins”, “Outs”, and Other Major Currencies in a Panel Cointegration Framework. IMF Working Papers 99/175. International Monetary Fund. Aykut, D., Sayek, S., 2007. The role of the sectoral composition of foreign direct investment on growth. In: Piscitello, Lucia, Santangelo, Grazia D. (Eds.), Do Multi-Nationals Feed Local Development and Growth? Elsevier, Amsterdam. Babecky, J., Bulír, A., Smídkov a, K., 2009. Sustainable Real Exchange Rates in the New EU Member States: Is Fdi a Mixed Blessing?. European Economy e Economic Papers 368 Directorate General Economic and Monetary Affairs (DG ECFIN), European Commission. , K., 2010. Sustainable Real Exchange Rates in the New EU Member States: what Did the Great Babecky, J., Bulír, A., Smídkova Recession Change?. IMF Working Papers 10/198 International Monetary Fund. Balassa, B., 1964. The purchasing-power parity doctrine: a reappraisal. J. Polit. Econ. 72, 584. Balke, N.S., Fomby, T.B., August 1997. Threshold cointegration. Int. Econ. Rev. 38 (3), 627e645. Egert, B., Halpern, L., MacDonald, R., 2006a. Equilibrium exchange rates in transition economies: taking stock of the issues. J. Econ. Surv. 20 (2), 257e324. Egert, B., Lommatzsch, K., Lahreche-Revil, A., 2006b. Real exchange rates in small open OECD and transition economies: comparing apples with oranges? J. Bank. Finance 30 (12), 3393e3406. Halpern, L., Wyplosz, C., 1996. Equilibrium Exchange Rates in Transition Economies. IMF Working Papers 96/125. International Monetary Fund. Harrod, R., 1933. International Economics. James Nisbet and Cambridge University Press, London. Johansen, S., Juselius, K., May 1990. Maximum likelihood estimation and inference on cointegrationewith applications to the demand for money. Oxf. Bull. Econ. Stat. 52 (2), 169e210. Kinoshita, Y., 2011. Sectoral Composition of Foreign Direct Investment and External Vulnerability in Eastern Europe. IMF Working Papers 11/123. International Monetary Fund. Lothian, J., Taylor, M.P., 2008. Real exchange rates over the past two centuries: how important is the Harrod-Balassa-Samuelson effect? Econ. J. 118 (532), 1742e1763. €svirta, T., 1988. Testing linearity against smooth transition autoregressive models. Biometrika Luukkonen, R., Saikkonen, P., Tera 75 (3), 491e499. Samuelson, P.A., 1964. Theoretical notes on trade problems. Rev. Econ. Stat. 46 (2), 145e154. Sarno, L., Taylor, M.P., 2002. The Economics of Exchange Rates. Cambridge University Press. Taylor, M.P., Sarno, L., 2001. Real exchange rate dynamics in transition economies: a nonlinear analysis. Stud. Nonlinear Dyn. Econ. 5 (3), 1. Ter€ asvirta, T., 1994. Specification, estimation, and evaluation of smooth transition autoregressive models. J. Am. Stat. Assoc. 89 (425), 208e218. Tsay, R.S., 1986. Nonlinearity tests for time series. Biometrika 73 (2), 461e466.
Contents lists available at ScienceDirect
Journal of International Money and Finance journal homepage: www.elsevier.com/locate/jimf
Real exchange rates and transition economies Gianna Boero a, Kostas Mavromatis b, Mark P. Taylor c, * a
Department of Economics, University of Warwick, UK Department of Economics and Econometrics, University of Amsterdam, The Netherlands c Warwick Business School, University of Warwick and Centre for Economic Policy Research, UK b
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 14 April 2015
In a number of empirical studies, transition economies have been shown to be subject to the Harrod-Balassa-Samuelson effect. This implies that the currencies of these countries have experienced a prolonged appreciation in real terms as their convergence proceeded. In this paper we find that a long-run relationship exists between the real exchange rate, productivity differentials, real interest rate differentials and the capital account for eight transition economies of Central and Eastern Europe, using monthly data over a period which extends from 1996 to 2013. We find that there are two sources of appreciation of the currencies of these countries, namely the Harrod-Balassa-Samuelson effect and the capital account effect, and argue that their significance depends on the type of investment received by the countries. While long-run foreign direct investment enhances productivity, porfolio investment leaves productivity unaffected, so our argument is that the larger foreign direct investment relative to portfolio investment, the greater the contribution of productivity in the determination of the real exchange rate. Moreover, we find that while the variables are linked by a linear long-run equilibrium relationship, adjustment towards equilibrium is nonlinear and is well represented by a smooth transition mechanism where the degree of equilibrium correction is a function of the sign and/or the size of the deviation from equilibrium. Interestingly, we find that a logistic smooth transition model fits well a larger number of countries, by allowing a different response of the real exchange rate to misalignments of different sign. © 2015 Elsevier Ltd. All rights reserved.
JEL classification: E52 F42 Keywords: Harrod-Balassa-Samuelson effect Real exchange rate Capital account Nonlinear models
* Corresponding author. Warwick Business School, University of Warwick, Coventry CV4 7AL, UK. Tel.: þ44 2476 524 534. E-mail address: [email protected] (M.P. Taylor). http://dx.doi.org/10.1016/j.jimonfin.2015.04.002 0261-5606/© 2015 Elsevier Ltd. All rights reserved.
24
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
1. Introduction The Harrod-Balassa-Samuelson effect, introduced by Harrod (1933), Balassa (1964) and Samuelson (1964), states that countries that have high productivity gains experience prolonged real appreciations of their currencies against the currencies of less productive countries. A vast empirical literature focusing either on transition or on industrialized economies supports this perspective. For example, Halpern and Wyplosz (1996), using panel data techniques for a number of transition economies find significant evidence in favour of the Harrod-Balassa-Samuelson effect at a relatively early stage of the transition process. More recently, though, there has been a growing recognition that the HarrodBalassa-Samuelson effect accounts only partly for the observed appreciation of the currencies of transition economies. A detailed overview of this literature, covering both theoretical and empirical aspects of recent contributions, including several additional explanations for the movements of real exchange rates in transition economies, can be found in Egert et al. (2006a). Two recent studies by Babecky et al. (2009, 2010) offer interesting insights on the interaction between foreign direct investment, net exports and sustainable real exchange rates in the new EU member states, and their findings are particularly relevant for the focus of our paper. Using a calibrated and simulated theoretical model, these studies conduct a sectoral analysis of foreign direct investment (FDI) and argue that transition economies where FDI is absorbed by the tradable sector are likely to benefit substantially from capital inflows and experience real currency appreciation. Building upon this literature, using time series econometric techniques, in this paper we take an innovative approach to investigate the way the composition of the capital account affects productivity and the real exchange rate in transition economies. We explore this channel by revisiting the behaviour of the real exchange rate in eight Central and Eastern European (CEE) countries, namely the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, the Slovak Republic and Slovenia, using monthly data over a sample period which goes from 1996 to the end of 2013. This period covers relatively early stages of the transition from planned to market-based economies, includes the entire period of preparation for accession to the European Union, which for these eight countries took place on 1 May 2004, and extends to the most recent years covering the Great Recession of 2008e09, which has clearly had an impact on the state of the convergence of these countries. We begin our empirical analysis by examining whether a long-run relationship exists among the real exchange rate, the productivity differential, the real interest rate differential and the capital account, and find supportive evidence of large productivity differential and capital account effects on the real exchange rate for a number of these transition economies. In addition, we show that the capital account effects depend on the composition of the country's capital inflows. In particular, we find that the capital account plays a role in the appreciation of the currency only to the extent that foreign direct investment exceeds portfolio investment. This is because FDI enhances productivity, while productivity is not affected by portfolio investment and, consequently, economies in transition can achieve a faster convergence by attracting long-run productivity-enhancing FDI. Long-term capital flows are the FDI which fosters growth, as it is generally taken to be determined by long-term profitability considerations and often leads to the transfer of state-of-the-art technology. As a result, these kinds of flows are less subject to market sentiment. On the contrary, portfolio investment and short-term bank lending (financial products) are short-run investments and subject to any kind of market volatility. Therefore, if portfolio investment is the key component of capital inflows to a transition economy, then it is likely that productivity is less strengthened compared to developed countries. Secondly, having established a long-run relationship between the real exchange rate and the variables mentioned above, namely, the productivity differential, the real interest rate differential and the capital account, we explore the possibility that the variables cointegrate in a nonlinear fashion, such that the speed of adjustment towards the long-run relationship varies over time, depending on the size and/or the sign of the exchange rate misalignment. A good deal of evidence in favor of nonlinear dynamical structure for the exchange rate and other macroeconomic variables has been reported in the literature. For example, in an application to transition economies during the 1990s, Taylor and Sarno (2001) find that the real exchange rate and the real interest rate differential cointegrate in a nonlinear fashion. Lothian and Taylor (2008) also employ models with nonlinear adjustment to test for
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
25
the effects of productivity differentials on the real exchange rate in developed economies and find cases where the variables cointegrate in one regime but not in the other. In line with the findings of these studies, our statistical analysis, based on a longer and more recent sample period for transition economies, provides strong evidence of nonlinear adjustments of the exchange rates in all countries under consideration. The remainder of the paper is organized as follows. In Section 2 we describe the environment in transition economies and set out the long-run equilibrium relationship which forms the basis of our subsequent analysis. In Section 3 we present the dataset and in Section 4 the econometric modelling strategy. In Section 5 we discuss the empirical results and Section 6 concludes. 2. The environment in transition economies After the initial recession, transition economies have been characterized by substantial productivity gains particularly in their industrial sector, coupled with a steady increase in the relative price of nontradables and a trend appreciation in the real exchange rate. So, the Harrod-Balassa-Samuelson (HBS) effect can be expected to be at work in these economies, and indeed, their currencies have been the subject of a large number of studies estimating the extent to which the HBS effect is reflected in real exchange rate movements. More precisely, in the literature, researchers assume that there are two sectors, tradables and non-tradables and that the increase in productivity takes place in the tradables sector, which consists of goods that are exported. A rise in productivity causes a rise in wages in the tradables sector, while prices remain unchanged due to perfect competition in the market for goods that are traded internationally. At the same time, the labour market is assumed to be perfectly competitive and labour is perfectly mobile within each country, but not across countries. Consequently, wages in the tradables and the non-tradables sectors are equalised. This will cause an increase in the price of non-tradables with a consequent rise in the overall price index. Assuming Purchasing Power Parity (PPP) holds in the tradables sector, the overall rise in the price level e at the prevailing nominal exchange rate e will result in an appreciation of the domestic currency in real terms. Before the start of the convergence process, transition economies experienced a depreciation of their currencies, sometimes quite abruptly, which was then followed by a systematic appreciation. One of the reasons for the appreciation of the currency was the increased labour mobility observed towards the more productive sectors, as inefficient production units shut down. Moreover, this appreciation of the real exchange rate has been further supported by inflows of direct investment. As productivity levels were converging to those of the industrialized countries, the rates of appreciation of the currencies started to slow down. This gradual phasing out of the appreciation of the real exchange rate experienced by the transition economies under consideration is illustrated in the plots in Fig. 1. For most countries, the real exchange rate appreciated steadily from the beginning of the sample to 2002, and then flattened considerably afterwards or even reversed its course, around the years of the Great Recession, turning into gradual depreciation in some of the countries. The real exchange rate is constructed, in logarithmic form, as follows:
q ¼ e þ p* p
(1)
where e is the logarithm of the nominal exchange rate, defined as the domestic price of one unit of foreign (Euro) currency, and p* and p are the logarithms of the foreign (Eurozone) and domestic price levels, respectively. Thus, an increase in the real exchange rate q corresponds to a depreciation, while a fall indicates an appreciation. Since, the purpose of the present paper is to examine whether the dynamics of the real exchange rate of transition economies are determined by factors other than the HBS effect, as the starting point of our analysis we consider the following long-run equation:
qt ¼ a0 þ a1 yt þ a2 rt rt* þ a3 CAt þ ut
(2)
where ut is an error term, yt is the logarithm of relative productivity (domestic productivity relative to foreign productivity), rt rt* is the real interest rate differential, where rt and rt* denote the domestic and foreign real interest rate respectively, and CAt denotes the capital account to GDP ratio. This relationship is very close to a standard specification based on the stock-flow approach to the real
26
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
Fig. 1. (Log) Real exchange rates.
exchange rate, advocated, for instance, by Alberola et al. (1999). According to the HBS effect, and other explanations focusing on the tradable-based real exchange rate, the coefficient on relative productivity is expected to be negative, that is, a rise in relative productivity causes a real exchange rate appreciation induced through relative price adjustment. On the other hand, New Open Economy Macroeconomics models show that productivity gains can translate into a real depreciation of the tradable-deflated real exchange rate through the terms-of-trade channel, and the overall impact on the real exchange rate of the whole economy will depend on which channel prevails (see Egert et al., 2006a; Egert et al., 2006b). In this paper, we argue that the composition of capital flows into the transition economies plays an important role in the explanation of the exchange rate dynamics of these economies. Below, we show that both the coefficient on the capital account and on relative productivity (HBS effect) depend on the type of investment which flows into these economies. During their transition process, these economies received large inflows of investment and, consequently, their capital accounts experienced long-lasting surpluses. However, not all capital inflows enhance productivity. The latter is increased with FDI, whereas portfolio investment (PI) does not contribute to productivity, and therefore capital account surpluses do not necessarily translate into higher productivity. Finally, as capital market integration is still an ongoing process in these economies, interest rate differentials may be an additional factor underlying the behavior of the real exchange rate, and is therefore included in our model. According to the traditional Mundell-Fleming-Dornbush model, real exchange rates and real interest rate differentials should be inversely related and move in opposite directions in the long run (see, e.g., Sarno and Taylor, 2002, ch 4).
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
27
3. The data We consider the following countries: the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, the Slovak Republic and Slovenia, and compare the first six countries with the Eurozone-17, while the last two are compared with the Eurozone-12.1 The data are taken at a monthly frequency from the International Monetary Fund International Financial Statistics, the Organization for Economic Cooperation and Development Main Economic Indicators and Eurostat. We use the end-of-period nominal exchange rate of the currency of each country against the Euro. For the period before the functioning of the Eurozone we use the ECU. The long-term government bond rate is used as a proxy for the nominal interest rate. The German 10-year government bond rate is used as a proxy for the long-term rate in the Eurozone. The consumer price index (CPI) is used as a proxy for the price level for each of the transition economies, while the harmonized index of consumer prices (HICP) is used as a proxy for the price level in the Eurozone. We measure productivity as the ratio of industrial production, excluding construction, to the number of people employed in industry. The capital account is measured as domestic assets held by foreign residents minus foreign assets held by domestic residents expressed in terms of GDP.2 Our sample starts from 1996 in order to eliminate the early years of transition during which productivity improvements were the result of initial reforms rather than the HBS effect, and spans through 2013.3 Both the start and end dates for each country is based on data availability. In particular, the sample periods used are the following: for the Czech Republic 1996:1e2013:2, for Estonia 1997:4e2011:7, for Hungary 1996:1e2013:7, for Latvia 1996:2e2013:9, for Lithuania 1997:1e2011:7, for Poland 1996:1e2013:9, for Slovenia and Slovak Republic 1996:1e2012:3. 4. Econometric modelling strategy The first step in our empirical strategy is to test whether Equation (2) constitutes a long-run relationship among the four variables considered. For this purpose we apply the DickeyeFuller and the Johansen test for cointegration. A finding of cointegration implies that the variables have an error correction representation, allowing for the long-run equilibrium relationship and at the same time capturing the short-run dynamics of the variables. A number of studies have reported evidence in favor of non-linear adjustments of the exchange rate towards its equilibrium value (see, amongst others, Taylor and Sarno, 2001, for transition economies, and Lothian and Taylor, 2008, for developed countries). If we find evidence of cointegration, then we explore the possibility that the mean reversion properties of the cointegrating residuals vary with the extent of the deviations from equilibrium. A suitable framework for modelling potential nonlinearities of this kind is the Smooth Transition Autoregressive (STAR) model. These models have received increasing attention in the econometric time series literature with applications to exchange rates and other macroeconomic variables.4 Our €svirta empirical analysis is conducted within this setting and follows the procedure suggested by Tera (1994). The procedure starts with the detection of nonlinearities in the cointegrating residuals of Equation (2), using the tests described in Luukkonen et al. (1988). The null hypothesis of linearity is tested against the alternative of an exponential smooth transition autoregressive (ESTAR) model or a logistic smooth transition autoregressive (LSTAR) model. These models describe, in different ways, a smooth transition between two extreme regimes for the adjustment of the real exchange rate. In the ESTAR model, the adjustment is symmetric, in the sense that only the distance from equilibrium matters, but not the sign. For small deviations from equilibrium, the model depicts the 'inner regime', while at the other extreme, for large misalignments above and below the equilibrium level, the model
1 Following the suggestion of an anonymous referee, we compare the Slovak Republic and Slovenia to the Eurozone (12 members) since Slovenia was the 13th member while Slovak Republic was the 16th. 2 Monthly GDP was obtained by linear interpolation from the quarterly series. 3 In an earlier version of the paper we included the years before 1996. However, following the suggestion of an anonymous referee, we have now excluded those initial years from our sample. 4 Taylor and Sarno (2001) found that the adjustment process of the real exchange rate in nine transition economies over the period leading to the end of 1997 was well described by an exponential smooth transition autoregressive process (ESTAR).
28
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
depicts the 'outer regime'. As the exchange rate moves further away from equilibrium, the model describes a nonlinear adjustment process with mean-reverting behavior which varies with the size of the deviation from equilibrium. The LSTAR model describes an asymmetric adjustment. The model parameters smoothly change between two extreme regimes, one characterized by large appreciations, the other by large depreciations, with one regime having more impact than the other according to the values of the misalignment. Formally, the STAR model for the cointegrating residuals ut is represented as follows:
u t ¼ c0 þ
Xp
cu i¼1 i ti
þ ðc00 þ
Xp
c0 u ÞGðutd ; k; cÞ þ et i¼1 i ti
(3)
This model consists of two autoregressive parts linked by a continuous nonlinear transition function, G(utd;k,c), bounded between zero and one, allowing for changes in the model parameters according to the value of the transition variable utd relative to a threshold value c. The transition function is exponential G(utd;k,c) ¼ 1 exp(k(utd c)2) in the ESTAR model, and logistic G(utd;k,c) ¼ (1 þ exp(k(utd c)))1 in the LSTAR model. These functions describe, in different ways, a smooth transition of ut between two extreme regimes associated with the extreme values of the transition functions. The parameter k > 0 is a transition parameters that governs the speed of transition between regimes, with high values implying a fast adjustment between regimes; et is an iid process; p is the lag-length, and d is a delay parameter. In the ESTAR model, for small deviations from equilibrium, as utd approaches c, the transition function approaches zero, and the model depicts the 'inner regime' where the series is described by a linear autoregressive model with coefficients c0 and ci, i ¼ 1,...,p. As utd moves further away, above or below the threshold value c, the model describes a nonlinear adjustment process with a mean-reverting behavior which varies with the size of the deviation from equilibrium. At the extreme, in the 'outer regime', for very large misalignments of both signs, the transition function approaches one, and the series is described by an autoregressive model with 0 0 intercept equal to c0 þ c0 and autoregressive coefficients equal to ci þ ci for i ¼ 1,...,p. Also, the model depicts a linear autoregression as k approaches zero or infinity, as the transition function becomes constant (equal to zero and one respectively). The LSTAR model describes an asymmetric adjustment toward c. The logistic transition function changes monotonically from zero to one as utd increases, and G(utd;k,c) ¼ 0.5 when utd is equal to c, the threshold between the two regimes. As in the ESTAR model, in the limit, as k approaches zero or infinity, the value of the transition function is constant, so the model becomes a linear autoregressive process. For intermediate values of k, the transition function approaches zero for values of utd far below c (large appreciations), and approaches one for values far above c (large depreciations). The model parameters smoothly change between these two extremes, with one regime having more impact than the other according to the values of utd c. The following reparameterization of model (3) is often used when studying the mean reversion properties of the series, when the true data generating process is an ESTAR or an LSTAR process:
Dut ¼ w0 þ wut1 þ
X p1 i¼1
wi Duti þ ðw00 þ w0 ut1 þ
X p1 i¼1
0
w0i Duti ÞGðutd ; k; cÞ þ et 0
(4)
0
This STAR formulation where w ¼ (c1 þ ... þ cp 1) and w ¼ ðc1 þ ::: þ cp 1Þ will display global 0 stability as long as ðw þ w Þ < 0. However, this condition does not rule out the possibility that the process is locally nonstationary, with w ¼ 0, implying unit root behavior in the linear part, or even the possibility of locally explosive dynamics, with w > 0. Following Luukkonen et al. (1988), a test for the null of linearity against the alternative of either an ESTAR or LSTAR stationary model can be computed using the auxiliary regression:
ut ¼ b1 zt þ b2 zt utd þ b3 zt u2td þ b4 zt u3td þ xt
(5)
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
29
0
where zt ¼ ½1; ut1 ; …; utp , the auxiliary parameters bi ¼ [bi0, bi1,..., bip], for i ¼ 1, 2, 3, 4, are functions of the parameters of the original STAR model, and xt is an error term.5 The test for the null hypothesis of linearity against STAR nonlinearity consists of testing the following restrictions:
HL : b2 ¼ b3 ¼ b4 ¼ 0 The order p of the linear autoregressive model is selected on the basis of the partial autocorrelation function of the series, and the delay parameter d for the transition variable utd is selected according to the procedure suggested by Ter€ asvirta (1994) as the value that produces the strongest rejection (the smallest p-value) of the null hypothesis of linearity. If the null hypothesis is rejected, the next step is to decide whether an ESTAR or an LSTAR is the appropriate model. The proposed decision rule consists of testing the following sequence of null hypotheses:
H01 : b4 ¼ 0; H02 : b3 ¼ 0 j b4 ¼ 0; H03 : b2 ¼ 0 j b4 ¼ b3 ¼ 0 The above sequence of tests can be performed as conventional Lagrange multiplier (LM) tests with an asymptotic chi-square distribution or using the F version of the LM test, which is preferred in small samples. If the p-value of the test corresponding to H02 is the smallest among the three tests, the ESTAR model should be selected, while in all other cases an LSTAR model is the appropriate choice. If nonlinearity in the cointegrating residuals is detected, then a nonlinear version of the error correction model, where the real exchange rate adjusts nonlinearly to past deviations from its long-run equilibrium, would provide a suitable description of its short-run dynamics. 5. Empirical results 5.1. The long-run equation In this section we present the results of the estimation of the long-run relationship (Equation (2)) and analyze the mean reverting properties of the residuals of this equation. In order to test the validity of Equation (2) as a long-run equilibrium relation, we start by testing whether the real exchange rate (qt), relative productivity (yt), interest rate differential ðrt rt* Þ and capital account (CAt) are cointegrated. Stationarity tests (available upon request) suggest that all four variables are integrated of order one; hence, we proceed to test for the existence of a common trend among them. We use both the DickeyeFuller test on the residuals of the long run regression and the Johansen cointegration procedure. The latter performs reasonably well in terms of power and size, when the adjustment towards equilibrium is nonlinear (see Balke and Fomby, 1997). We computed both the maximum eigenvalue test and the trace test and obtained similar results, therefore, in Table 1 (Panel A) we report only the results from the former test. While the DickeyeFuller tests were unable to reject the null hypothesis of non-stationarity in most cases, the existence of one cointegrating vector could be established with the Johansen test for all countries either at the 1-percent or at the 5percent significance levels. The maximum likelihood estimates of the cointegrating coefficients, based on the VAR specification, are reported in Panel B of Table 1. Any potential endogeneity of the variables in the model is taken into account by the Johansen approach used to estimate the long-run coefficients. This means that we account for issues of potential simultaneity that could arise, for example, from productivity being determined by the capital account and vice-versa. Therefore, the VAR approach is particularly suitable given that productivity gains have been driven by large capital inflows in these countries, and these two variables are potentially strongly correlated. Essentially, by considering the CAt variable in the long-run equation together with the productivity variable, we control for the potential effects of capital flows on productivity, and therefore the coefficient of CAt can be interpreted as an estimate of its impact
5
The auxiliary regression is based on a third-order Taylor series approximation of the transition function.
30
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35 Table 1 Cointegration results. Panel A: Cointegration test statistics
Czech Republic Estonia Hungary Latvia Lithuania Poland Slovak Republic Slovenia
D-F test
Joh. test: max (r ¼ 0 vs r ¼ 1)
2.13710 2.87483 3.14199* 1.83711 1.40915 2.83779 2.35878 2.45255
35.0052*** 28.8818** 32.6793** 59.4450*** 31.7927** 40.3253*** 47.8784*** 29.2182**
Panel B: Estimated cointegrating vectors
Czech Republic Estonia Hungary Latvia Lithuania Poland Slovak Republic Slovenia
yt
(r r*)t
C At
Constant
0.849 1.668 0.255 1.231 0.259 0.195 0.400 0.569
0.051 0.060 e 0.105 0.017 0.189 0.036 0.070
0.860 0.034 0.083 0.341 0.040 0.199 0.010 0.024
3.489 0.598 e 0.408 0.135 0.943 e 0.292
Notes: Panel A:***Significant at 1%. **Significant at 5%. *Significant at 10%. For the Johansen test (Joh. Test), the critical values for the max test statistics, for H0: r ¼ 0 vs r ¼ 1, are 33.24 for the 1%, 28.14 for the 5% and 25.56 for the 10% significance level, respectively (Johansen and Juselius (1990)). Panel B: Maximum Likelihood estimates. The cointegrating vectors are derived from the Johansen procedure and normalized so that to receive the parameter values corresponding to Equation (2).
on the real exchange rate, once the indirect effect on productivity has been taken into account. The estimation results in Table 1 show a negative coefficient for productivity for all countries, implying that the currencies of these countries are subject to the HBS effect. As the catching-up process of those economies continues, increases in relative productivity induce an appreciation of the real exchange rate. This result is in line with the recent literature. For example, in a panel study for 11 transition economies, Egert et al. (2006b) provide evidence in favor of a negative effect of productivity on the real exchange rate.6 Table 1 also shows that the value of the coefficient on relative productivity differs substantially across countries, revealing that the HBS effect is more important for some countries (Czech Republic, Estonia, Latvia, Slovak Republic and Slovenia) than for others (Hungary, Lithuania and Poland). The real interest rate differential has a negative sign except for Poland, Slovak Republic and Slovenia, and the coefficient of the capital account variable is negative for all countries, except for Hungary Lithuania and Poland. As real exchange rate appreciation is the equilibrium behavior of the exchange rate of these economies over most of the period considered, the currencies of the countries where the coefficient on the capital account is negative will have two sources of appreciation: the productivity differential and the capital account surplus (or the current account deficit). An examination of how FDI and PI evolved through time in each country provides useful insights for the interpretation of the results obtained from the estimation of Equation (2). Our findings are in line with some of the recent literature which has investigated the direct link between FDI and productivity. For example, Babecky et al. (2009) conduct an analysis based on a structural medium-term model of sustainable real exchange rates and argue that inflows of foreign direct investment tend to appreciate the currencies of those
6 Their sample, though, spans until 2004, while their emphasis is on net foreign assets, rather than on the capital account as a whole. Moreover, they use the real effective exchange rate for each country instead of comparing the currencies of those economies to the Euro as is the case in this paper.
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
31
economies in real terms, as long as they are absorbed by the tradable goods sector.7 Our results suggest that for the countries with a negative coefficient on the capital account, the effect of productivity (the HBS effect) is accentuated with respect to those countries where the capital account surpluses cause real exchange rate depreciation. A further examination of the effects of the capital account conducted in the next section suggests that the coefficient on productivity may depend on the kind of investment that each country receives. In particular, we argue that for countries with high levels of FDI relative to portfolio investment, productivity proves to be the main driving force behind the prolonged real appreciation of their currencies, while this is not the case when FDI and portfolio investment are equally important. 5.2. What explains the long-run coefficient on the capital account and productivity? In order to determine the importance of the capital account in the determination of the real exchange rate in the transition economies, we examine the evolution of FDI and portfolio investment (PI) over the period considered. While FDI enhances productivity, PI leaves productivity unaffected. We collected data on net inward FDI and PI for each economy as a percentage of GDP. For countries in which FDI represents, on average, a substantially higher percentage of GDP than PI, the coefficient on the capital account in the long-run Equation (2) exhibits a negative value. This is the case for the Czech Republic, Latvia, the Slovak Republic and Slovenia, and these are also the countries with a higher negative coefficient on productivity differentials. This means that these countries have received important waves of labour augmenting productivity investment. For example, for the Czech Republic, FDI amounts to 4.57% of GDP and PI only to 0.40% on average over the sample period, and for the Slovak Republic, FDI is 3.28% of GDP and PI only 0.38%. On the other hand, for Poland, Hungary and Lithuania, FDI and PI are much closer, on average, and the effects of FDI are likely to be offset by those of PI. For Poland, for example, FDI amounts to 2.8% of GDP and PI to 2.28%, for Hungary, FDI amounts to 2.66% and PI to 1.76%, while for Lithuania, FDI amounts to 3.58% and PI to 4.94%. For these three countries, the capital account coefficient in the long-run relationship is positive, in contrast to the other countries; moreover, the coefficient on the productivity differential is lower than in the other countries. Overall, this analysis supports our conjecture that high levels of FDI, relative to portfolio investment, strengthen the effect of productivity on the real exchange rate, alongside the usual HBS effect. On the other hand, when FDI and PI are equally important, then financial factors play a significant role in the determination of the real exchange rate, which may be reflected in a reduced long-run effect of productivity. Previous studies based on a decomposition of FDI have shown that while FDI in the tradables sector leads to an improvement of the external balance, creating economy wide externalities, when FDI inflows are mainly absorbed by the non-tradable goods and services sectors, then such inflows could be associated with a deterioration of the trade balance, which would require a real exchange rate depreciation in order for the economy to return to a sustainable path of economic development (see Kinoshita, 2011; Babecky et al., 2009, 2010; Aykut and Sayek, 2007). In such a case, relative productivity will have either a weak negative effect on the real exchange rate (i.e. appreciation) or a positive effect (i.e. depreciation). In order to explore this issue further, for each country we constructed indices of productivity for the manufacturing (tradables) and construction (non-tradables) sectors and compare their evolution over time. As shown in Fig. 2, for Hungary, Lithuania and Poland, productivity in the construction sector is higher relative to that in the manufacturing sector during most of the sample period considered, which may reflect the fact that FDI is absorbed to a higher proportion by the nontradables sector, and may be taken as an additional factor explaining the estimated positive capital account coefficient for these countries. Thus, when FDI inflows are mainly absorbed by the nontradable goods and services sectors, such inflows could be associated with a deterioration of the trade balance, which would require a real exchange rate depreciation in order for the economy to
7 On the other hand, Egert et al. (2006b) show that an increase in net foreign liabilities implies a real exchange rate depreciation for the countries considered in our paper with the exception of Latvia. However, their analysis is conducted in a different context as they consider an earlier period, which starts in 1973, and use the real effective exchange rate as the variable of interest.
32
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
Fig. 2. Productivity in manufacturing and construction industry.
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
33
Table 2 Linearity of the cointegrating residuals.
Czech Republic Estonia Hungary Latvia Lithuania Poland Slovak Republic Slovenia
HL (d)
H01
H02
H03
0.0175( b d ¼ 14) 0.0103 ( b d ¼ 8) 0.0000 ( b d ¼ 6) 0.0112 ( b d ¼ 3) 0.0212 ( b d ¼ 3) b 0.0254 ( d ¼ 3) 0.0059 ( b d ¼ 8) 0.0028 ( b d ¼ 6)
0.6909 0.2783 0.0000 0.5204 0.0068 0.0230 0.2820 0.4264
0.0072 0.0370 0.0261 0.0022 0.2378 0.0436 0.1039 0.0134
0.0968 0.0162 0.4891 0.2459 0.3265 0.8274 0.0032 0.0065
Notes: P-values of the LM test statistics described in Section 4. If the P-value of the test corresponding to H02 is the smallest the ESTAR model is selected, otherwise the LSTAR model is the appropriate choice.
return to a sustainable path of economic development. At the same time, if the economy receives equally high portfolio investment (as is the case with Hungary, Lithuania and Poland), it needs a trade surplus in order to be able to pay back its external debt.8 Unless FDI inflows go to tradables, the economy would need a real exchange rate depreciation in order to satisfy this requirement. 5.3. Nonlinear short-run dynamics In this section we present the results of an investigation of the short-run dynamics of the exchange rates for the eight transition economies considered. Having found a long-run equilibrium relationship, we explore the possibility that the short-run adjustment towards equilibrium varies with the extent of the deviations from equilibrium. In particular, we explore the existence of potential STAR nonlinearity in the cointegrating residuals of Equation (2), following the procedure described in Section 4. The results of the linearity tests for each country and the delay parameter of the transition variable, d, are presented in Table 2. The null hypothesis of linearity against a STAR alternative is rejected at the 5% or 1% level of significance for all countries, implying the existence of nonlinear behavior of the short-run adjustment process. The delay parameter in some cases is high (d ¼ 14 for the Czech Republic, and d ¼ 8 for Estonia and the Slovak Republic), reflecting a slow response to shocks in the real exchange rate of these countries. For most countries, the smallest p-value is found for either H01 or H03, and this result, €svirta decision rule, suggests an LSTAR specification. The only two countries for according to the Tera which the smallest p-value occurs for H02, suggesting an ESTAR specification, are the Czech Republic and Latvia. As explained in Section 4, LSTAR nonlinearity implies an asymmetric adjustment of the cointegrating residuals, depending on both the size and the sign of past disequilibria, while ESTAR nonlinearity implies a symmetric adjustment for deviations above and below the equilibrium level. We then fit ESTAR or LSTAR models to the cointegrating residuals of each country, according to the results of the linearity tests. The models are estimated using the reparameterization (4) described in Section 4. In Table 3 we report the estimates of the coefficients w and w', providing information on the degree of mean reversion across regimes, with p-values in brackets. All models show global stability, 0 that is ðw þ w Þ < 0. This is not surprising, as we would expect cointegrating residuals to be globally stationary. However, there are cases where the process is locally nonstationary, indicating that adjustment to the long-run equilibrium occurs in one regime, but not in the other. For example, the ESTAR model for the Czech Republic indicates non-mean reversion (lack of cointegration) in the inner regime, that is, in the neighborhood of the equilibrium, and mean reversion (the variables cointegrate) in the outer regime, characterized by large misalignments of both signs, a result which is consistent with previous studies. Also interestingly, the LSTAR models for Lithuania and for the Slovak Republic show non-mean reversion in the extreme regime characterized by large over-valuations, and mean reversion in the other extreme regime characterized by large under-valuations. The remaining results
8 Further research might usefully examine the composition of portfolio flows in terms of equity and debt, and whether this has a bearing on this analysis.
34
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
Table 3 Estimated STAR models for the cointegrating residuals. w Czech
e
Estonia
0.177 (0.010) 0.112 (0.441) 0.194 (0.031) 0.071 (0.167) 0.106 (0.031) 0.182 (0.113) 0.671 (0.000)
Hungary Latvia Lithuania Poland Slovak Republic Slovenia
w0
kE
0.184 (0.083) 0.163 (0.097) 0.194 (0.090) 0.193 (0.017) 0.500 (0.000) 0.638 (0.000) 0.524 (0.024) 0.462 (0.002)
0.800 [0.023]
kL
0.936 [0.010] 0.871 [0.015] 0.923 [0.034] 0.144 [0.023] 2.339 [0.035] 0.650 [0.003] 0.721 [0.019]
R2
LMARCH
TS
0.176
{0.367}
{0.175}
0.116
{0.411}
{0.207}
0.167
{0.064}
{0.159}
0.181
{0.504}
{0.081}
0.121
{0.133}
{0.802}
0.096
{0.849}
{0.466}
0.160
{0.459}
{0.113}
0.288
{0.529}
{0.904}
Notes: P-values in parentheses. kE and kL are the transition parameters for the ESTAR and LSTAR model respectively, and the numbers in square brackets are the empirical P-values obtained from Monte Carlo simulations. The last two columns report the P-values for the LM test for autoregressive conditional heteroskedasticity and the Tsay test (TS) for neglected nonlinearities (Tsay (1986)).
for the other countries show that the real exchange rate adjustment in all cases is highly nonlinear, and the type and speed of adjustment is country specific, reflecting different reactions to correct misalignments of different sign and size. Table 3 also reports the estimated transition parameters kE (for the ESTAR models) and kL (for the LSTAR models), implying a moderate speed of transition between regimes for most countries, the highest being in the LSTAR model for Poland (kL ¼ 2.34). The last two columns of Table 3 report diagnostic tests showing that the estimated models perform satisfactorily. Our finding of LSTAR nonlinearity for the majority of the transition economies considered in this study is new, and suggests a different dynamics from that found in previous studies, which have reported evidence of symmetric ESTAR nonlinearity (see, for example, Taylor and Sarno, 2001). LSTAR models allow the exchange rate to respond differently not only to the size of the deviations from equilibrium, but also to over-valuations and under-valuations. This type of asymmetry seems more plausible over a period of such significant changes for these economies which have been subject to pressures of various kinds at different stages of their transition process, to the challenges posed by accession to the EU, and more recently by the Great Recession. Our results are indicative of the complexity of the economies of these countries, and LSTAR nonlinearity may better reflect policy interventions which react differently to correct misalignments of different size and sign. A full exploration of the short-run dynamics, allowing also for the possibility of endogeneity among all of the four variables considered in this studydsuch as the extent to which productivity gains are driven by capital inflows, for exampled, would require the use of a multivariate smooth transition equilibrium correction model. However, this analysis goes beyond the scope of the present paper, and represents an avenue for future research. 6. Conclusion The behavior of the real exchange rate in transition economies, particularly during the transition and catching-up process, has been the focus of several studies. In this paper we have reexamined the behavior of the real exchange rate in eight transition economies of Central and Eastern Europe, using a data set which extends from the early period of sustained appreciation in the mid 1990s, to the most recent period, characterized by a gradual phase-out of the real appreciation. The empirical contribution of this paper is twofold. First, we showed that there are two sources of appreciation of the real exchange rate: the HBS effect, via productivity differentials, and capital account effects. In particular, we argued that the HBS effect depends itself on the composition of the capital flows in each country: high levels of FDI relative to portfolio investment strengthen the effect of productivity, and hence the HBS
G. Boero et al. / Journal of International Money and Finance 56 (2015) 23e35
35
effect on the real exchange rate is accentuated. On the other hand, the HBS effect is weakened in countries where foreign direct investment are on average, over the sample period, close to portfolio investment; in this case, financial factors are equally important in determining the real exchange rate, thereby reducing the net effect of productivity. We then showed that while the variables are linked by a linear long-run equilibrium relationship, the adjustment of the real exchange rate towards equilibrium is nonlinear and is well described by a smooth transition mechanism where the degree of equilibrium correction is a function of the sign and/ or the size of the deviation from equilibrium. Interestingly, we find that for most countries a logistic smooth transition model fits well the short run dynamics of the real exchange rate, allowing a different response to misalignments of different sign. Our finding of LSTAR nonlinearity for the majority of the transition economies considered in this study is new, and suggests a different dynamics from that found in previous studies, which have reported evidence of symmetric ESTAR nonlinearity. LSTAR models allow the exchange rate to respond differently not only to the size of the deviations from equilibrium, but also to over-valuations and under-valuations. This type of asymmetry may reflect policy interventions which react differently to misalignments of different sign, and seems plausible for economies which have been exposed to shocks of various kinds at different stages of their transition process, to the challenges posed by accession to the EU, and more recently by the Great Recession. Now, more than ten years after joining the EU (May 2004), these economies are still in the process of catching up to the EU level of development. Acknowledgments The authors are grateful for constructive comments on an earlier version of this paper from the Editor, Menzie Chinn, and from an anonymous referee; any errors that remain are the responsibility of the authors alone. References Alberola, E., Cervero, S.G., Lopez, J.H., Ubide, A.J., Dec. 1999. Global Equilibrium Exchange Rates e Euro, Dollar, “Ins”, “Outs”, and Other Major Currencies in a Panel Cointegration Framework. IMF Working Papers 99/175. International Monetary Fund. Aykut, D., Sayek, S., 2007. The role of the sectoral composition of foreign direct investment on growth. In: Piscitello, Lucia, Santangelo, Grazia D. (Eds.), Do Multi-Nationals Feed Local Development and Growth? Elsevier, Amsterdam. Babecky, J., Bulír, A., Smídkov a, K., 2009. Sustainable Real Exchange Rates in the New EU Member States: Is Fdi a Mixed Blessing?. European Economy e Economic Papers 368 Directorate General Economic and Monetary Affairs (DG ECFIN), European Commission. , K., 2010. Sustainable Real Exchange Rates in the New EU Member States: what Did the Great Babecky, J., Bulír, A., Smídkova Recession Change?. IMF Working Papers 10/198 International Monetary Fund. Balassa, B., 1964. The purchasing-power parity doctrine: a reappraisal. J. Polit. Econ. 72, 584. Balke, N.S., Fomby, T.B., August 1997. Threshold cointegration. Int. Econ. Rev. 38 (3), 627e645. Egert, B., Halpern, L., MacDonald, R., 2006a. Equilibrium exchange rates in transition economies: taking stock of the issues. J. Econ. Surv. 20 (2), 257e324. Egert, B., Lommatzsch, K., Lahreche-Revil, A., 2006b. Real exchange rates in small open OECD and transition economies: comparing apples with oranges? J. Bank. Finance 30 (12), 3393e3406. Halpern, L., Wyplosz, C., 1996. Equilibrium Exchange Rates in Transition Economies. IMF Working Papers 96/125. International Monetary Fund. Harrod, R., 1933. International Economics. James Nisbet and Cambridge University Press, London. Johansen, S., Juselius, K., May 1990. Maximum likelihood estimation and inference on cointegrationewith applications to the demand for money. Oxf. Bull. Econ. Stat. 52 (2), 169e210. Kinoshita, Y., 2011. Sectoral Composition of Foreign Direct Investment and External Vulnerability in Eastern Europe. IMF Working Papers 11/123. International Monetary Fund. Lothian, J., Taylor, M.P., 2008. Real exchange rates over the past two centuries: how important is the Harrod-Balassa-Samuelson effect? Econ. J. 118 (532), 1742e1763. €svirta, T., 1988. Testing linearity against smooth transition autoregressive models. Biometrika Luukkonen, R., Saikkonen, P., Tera 75 (3), 491e499. Samuelson, P.A., 1964. Theoretical notes on trade problems. Rev. Econ. Stat. 46 (2), 145e154. Sarno, L., Taylor, M.P., 2002. The Economics of Exchange Rates. Cambridge University Press. Taylor, M.P., Sarno, L., 2001. Real exchange rate dynamics in transition economies: a nonlinear analysis. Stud. Nonlinear Dyn. Econ. 5 (3), 1. Ter€ asvirta, T., 1994. Specification, estimation, and evaluation of smooth transition autoregressive models. J. Am. Stat. Assoc. 89 (425), 208e218. Tsay, R.S., 1986. Nonlinearity tests for time series. Biometrika 73 (2), 461e466.