Exchange Rate Volatility Effect on Economic Growth under Different Exchange Rate Regimes: New Evidence from Emerging Countries Using Panel CS-ARDL Model

Abstract

This paper analyzes the impact of exchange rate volatility on economic growth under various exchange rate regimes. It empirically examines this issue in 14 emerging countries from 1990 to 2020. This study has three particularities: First, we use the GARCH model to generate the conditional variance, which will be used as a proxy variable for the exchange rate volatility. Second, to address our issue, we employ the Panel CS-ARDL model, one of the most recent models for handling panel cases. Third, we apply the Dumitrescu and Hurlin Granger non-causality test to capture the potential indirect effect that exchange rate volatility can have on economic growth through the channel of its determinants. The results of our study demonstrate that exchange rate volatility costs emerging countries both directly and indirectly in terms of growth. However, by controlling our countries according to the adopted exchange rate regime, we find that the magnitude of this impact tends to be stifled in the case of countries adopting intermediate exchange rate regimes. Through their combination of rigidity and flexibility, intermediate exchange rate regimes appear to be more effective in mitigating the direct effects of exchange rate volatility on economic growth.

1. Introduction

In the medium and long term, rapid economic growth and sustainable economic development are still the main macroeconomic objectives for emerging countries. Even though these economies have experienced successful growth over the past twenty years, the desired growth level has not always been achieved. According to Duval and Furceri (2019), emerging countries at their actual growth rates will have to wait more than fifty years to close half their income gap. In contrast, simultaneous reforms have to be implemented to accelerate the convergence of an emerging country’s income to the standard of living of an advanced country. However, Johnson et al. (2010) have pointed out that “there is no magic enabler for growth.” At least three main growth-destroying factors can be agreed upon, notably: the weaknesses of institutions; exposure to different types of conflict, which determines the country’s political stability (ethnic conflicts, geopolitical conflicts, etc.); and bad macroeconomic policy decisions. Since exchange rate policy is one of the pillars of macroeconomic policy, countries are expected to put it at the service of development by choosing the optimal exchange rate regime that allows them to avoid the risk of exchange rate misalignment.

The exchange rate regime is defined as the choice made by the authorities regarding the particular behavior of its exchange rate against other currencies over a given time horizon (Elbadawi and Kamar 2006). However, an inadequate choice can lead to dramatic and unexpected exchange rate movements with real economic and financial damage. The financial turmoil in Mexico in 1994, the currency crises in Asia from 1997 to 1998, and the rise in international financial imbalances from the 2008 crisis onwards are shreds of evidence of an inappropriate choice of exchange rate regime consequences.

Many authors focus on the internal factors determining the optimality of each kind of exchange rate regime in promoting economic growth. These works are inspired by the “theory of optimal currency areas” developed by Mundell (1961), which is an indispensable theory in the analysis of the choice of exchange rate regimes. According to his findings, the main benefit of fixing the exchange rate regime is reducing uncertainty. A fixed exchange rate system prevents exchange rate fluctuations, which eliminates the risk premium. Consequently, investors are reassured by the resulting stable environment, which has a favorable effect on investment behavior and, as a result, economic growth. However, the fixed exchange rate regime imposes a high cost on the country regarding the loss of monetary policy autonomy. When exchange rates are fixed, it becomes more difficult to rapidly influence the relative price evolution by altering nominal exchange rates. As a result, any adjustments in real rates that may be necessary must be made through relative wages and prices. However, if these are rigid, the adjustment process becomes more extended, which penalizes the country’s economic growth. Also widely accepted is the Balassa–Samuelson hypothesis, which attempts to explain exchange rates by international trade and economic development. The Penne effect explains why increases in productivity in the trade sector tend to exceed those in the non-trade sector. Thus, due to the productivity gains effect, tradable product prices are expected to rise. As a result, exchange rates, based on the CPI, rise in countries with rapid economic development. In contrast, they depreciate in low-growth countries (Balassa 1964; Samuelson 1964).

The lack of consensus on this issue has fueled our interest in analyzing the nature of the impact of exchange rate volatility on economic growth in emerging economies. Furthermore, we analyze the role of exchange rate regimes in protecting economic growth by absorbing exchange rate shocks. The paper will be conducted in two parts: theoretical and empirical. In the first part, in Section 2 we will focus on the controversy around how the exchange rate regime choice and the associated exchange rate volatility affect economic growth. After that, the empirical part will begin with Section 3, in which we will determine the empirical methodology followed in our study carried out on a panel of 14 emerging countries (Morocco, Egypt, Turkey, Brazil, China, South Africa, Thailand, Tunisia, Nigeria, Indonesia, Mexico, Jordan, Peru, and Bolivia) during the period 1990–2020. Section 4 will present the GARCH (1,1) modeling of REER volatility and the empirical validation of the various CS-ARDL Panel model development steps. In the last empirical Section 5, we will provide the answers to our issue by analyzing the CS-ARDL panel model results. However, we will refer to the Dumitrescu and Hurlin Granger non-causality test as a complementary tool to examine the indirect impact of the exchange rate on economic growth via its determinants.

2. Literature Review

The impact of the exchange rate regime and associated movements in the exchange rate on economic growth has been and will remain a recurrent topic of debate. The first generation of authors has refuted the existence of any dependence of countries’ economic growth on their exchange rate regime’s choice. This is the case for Baxter and Stockman (1989) and Baxter (1991) studies on a panel of 49 panels over two study periods (before and after the Bretton Woods system), and later Ghosh et al. (2003). However, their findings received much criticism on different levels: The authors first examined this causality in a unidirectional sense. Moreover, they disregard the classification of countries according to their level of development, which is a crucial factor in investigating this topic. Emerging economies, for example, suffer significantly from internal distortions that try to stifle economic growth (Miles 2006). This limit became the guiding thread for several further works.

Rogoff et al. (2003) examined 160 countries from 1940 to 2001 after categorizing them as emerging, developing, or developed. They discovered that the relationship between the exchange rate regime and economic growth in the first type of country is ambiguous. However, the floating exchange rate regime stimulates economic growth in developed countries. These theoretical observations have created fertile ground for extensive discussion. This ambiguity is due to the nature of the link between the exchange rate regime and economic growth. The impact is not always direct but can be channeled through economic growth determinants like investment, inflation, financial system development and depth, and external debt volume. For this reason, it is necessary to handle this issue from both direct and indirect perspectives.

The direct impact of the exchange rate regime on economic growth is explained by the role of the exchange rate as a shock absorber. Therefore, the exchange rate regime appears to be a determining factor in the international transmission of shocks. (Flood and Marion 1982; Baxter and Stockman 1989). Proponents of this view take Milton Friedman’s (1953) work as their starting point. In his essay on the case of flexible exchange rate regimes, he argues that flexible exchange rate regimes are more relevant in protecting the economy against internal and external shocks through the option of absorption and adaptation that they confer. He emphasizes that adjustment is mediated by changes in relative price levels in a rigid exchange rate regime. Nevertheless, this operation is slow in a Keynesian price world, which burdens the economy excessively and ultimately harms growth. In the same context, other authors concluded that the type of shock must be considered when analyzing this causality1 (Khan et al. 1991; Frankel 2003). As for real shocks, Rodney (2007), Ferrari-Filho and De Paula (2008), and Guzman et al. (2018) argue for the effectiveness of flexible exchange rate regimes for countries experiencing terms of trade shocks and nominal rigidities. In a fixed exchange rate regime, external shocks must be absorbed in other markets via different mechanisms (price variations to avoid a potential drop in production, wage flexibility, and worker mobility to avoid unemployment), making this option more costly. However, for domestic monetary shocks largely caused by money supply changes or velocity changes, Daly (2007) found that the fixed exchange rate regime appears to be the most effective in reducing spillovers. Its stabilizing role appears to moderate the consequences of rising demand for foreign products due to rising money demand. In the case of shocks from external sources, exchange rate flexibility is more suitable. The sensitivity of domestic interest rates to those of the anchor countries returned to the discussion after the rise in US interest rates in 1999–2000 (tightening period) for countries adopting a fixed exchange rate regime. Although exchange rate fixity provides a buffer against domestic monetary shocks and a nominal anchor for monetary policy credibility, during times of international market disruptions, exchange rate flexibility allows the country to preserve monetary policy autonomy (Frankel 1999; Frankel et al. 2002; Cuaresma and Wójcik 2006; Giovanni and Shambaugh 2008). Also, Obstfeld et al. (2018) reported that, particularly in emerging economies2, the transmission of international financial conditions to domestic macroeconomic variables is amplified under a fixed exchange rate regime.

As previously stated, the impact of the choice of the exchange rate regime can be mediated through the determinants of economic growth that are sensitive to the exchange rate regime choice and the associated exchange rate volatility. Investment, inflation, the volume of external debt, the depth of the financial system, the degree of trade openness, and the degree of financial system development are all factors to consider.

Many theoretical findings suggest that the exchange rate regime impacts economic growth via investment (Ramey and Ramey 1995; Aizenman 1992). Since an investment promotion strategy requires a stable economic environment, the uncertainty associated with exchange rate volatility under flexible exchange rate regimes may be sufficient to increase risk aversion among investors and curb investment consequently (Asteriou and Price 2005; McKinnon and Schnabl 2003). For portfolio investments, Dumas and Solnik (1995) and De Santis and Gerard (1998) find within the framework of international asset pricing theory that since exchange rate volatility is unpredictable, it embodies an additional risk that investors must take and for which they should be compensated. In other words, investors expect a risk premium on their investments due to the increased risk exposure caused by exchange rate fluctuations (Mahapatra and Bhaduri 2019). In this sense, Farhi and Werning (2012) pointed out that capital controls can be optimal even if the exchange rate is not fixed in response to risk premium shocks. The second category of investment, namely FDI, is no exception. Identically, Cushman and De Vita (2017), in their study of 70 developing countries from 1981–2013, found that exchange rate fixity is an incentive for FDI through the stability it generates for investors. In contrast, exchange rate volatility creates uncertainty for investors, which penalizes FDI inflows (Perekunah 2020).

Another significant element in this issue is the development of the financial markets as a contributor to the economy’s effectiveness (Levine 1997). Using panel data, Aghion et al. (2009) demonstrated that the long-term benefits of an exchange rate regime on economic growth only occur when the financial development level is considered. The Shaw (1973) reflection and McKinnon (1973) paradigm sparked this discussion about the significance of financial market development. They contend that the financial system plays a critical role in raising the level of savings, which is necessary for investment and growth. Bekaert et al. (2005) have shown that a developed financial environment makes it easier for investors to minimize risk through financial markets, lowers the cost of capital, stimulates investment, and leads to economic growth.

The third channel on which we will focus is the volume of external debt, whose management has been recognized as a major challenge for emerging countries for at least two reasons: On the one hand, it leads to a transfer of wealth to foreign lenders. On the other hand, the repayment ability of debt service is based on the foreign currency resources of the country concerned. The foundation for studies on this issue is laid in the writings of Edwards (1984) and Cline (1985). Gertler et al. (2007) and Cihak et al. (2012) have argued that the choice of exchange rate regime significantly impacts the external debt of developing countries. In particular, under a fixed exchange rate system, there is a higher probability that the nation will issue foreign bonds. Therefore, the country’s debt structure thus appears to be a determinant of the impact of the choice of exchange rate regime on economic growth (Eichengreen and Hausmann 1999; Bailliu et al. 2002), because when external debt represents the lion’s share of the country’s total debt, the flexible exchange rate regime becomes more beneficial for debt financing.

The impact of trade openness is theoretically inescapable and empirically proven to be a growth promoter (Edwards and Lederman 1998; Keho and Wang 2017). Thus, the degree of openness may embody a transmission channel for the impact of the exchange rate regime on economic growth. Rose (2000) suggests that sharing the same currency with the trading partner reduces transaction costs by eliminating the uncertainty of exchange rate fluctuations. Consequently, this unification reflects positively on bilateral trade between the two countries. Thus, the fixed exchange rate regime will be more profitable if the anchor currency is the same as its trading partner. Fritz-Krockow and Jurzyk (2004) similarly confirmed that defining the anchor currency of the exchange rate regime is crucial to maximizing the benefits of bilateral trade. Similarly, several authors have shown the detrimental effect of exchange rate volatility under flexible exchange rate regimes on countries’ trade flows. Arize et al. (2000), in their study on 13 less developed during the period 1973–1996, reveal that REER volatility has a considerable detrimental impact on export flows. Likewise, Olimov and Sirajiddinov (2008) reached the same results in their study on Uzbekistan from 2001 to 2003.

Our literature survey concludes with a discussion about the last channel transmission: “inflation.” Its impact on economic growth has been the subject of a broad but controversial theoretical and empirical debate. However, most authors have concluded that inflation does hurt growth, but only if it is above a specific threshold. In this context, Mubarik and Riazuddin (2005) examined the case of Pakistan and arrived at a threshold of 9%, above which inflation harms economic growth. Munir et al. (2009) to a threshold of 3.89%. Later, Akgül and Özdemir (2012) provided a maximum tolerable inflation rate of 1.29%. All these findings lead us to study the possibility that inflation might be a transmission channel for the effects of the exchange rate on economic growth. The “law of one price” and the “theory of purchasing power” serve as the foundation for analyzing the degree of exchange rate shocks on domestic prices (the degree of pass-through). Both show a perfect transmission (complete pass-through) of exchange rate shocks to prices. However, several empirical results later refuted these findings, leading to a partial pass-through in most economies, with the highest degree of pass-through found in emerging economies (Ito et al. 2005). This discrimination between these two kinds of transmission is of utmost importance for at least two reasons. First, from a commercial perspective, it makes it possible to predict the competitiveness of firms in international markets. Furthermore, from a second political angle, it allows us to anticipate the degree to which a depreciation policy can be an effective instrument for reducing the trade deficit. A market with a low pass-through, for example, requires a more considerable depreciation than a market with a complete or strong pass-through to reduce the trade deficit by the same level.

Our literature overview has allowed us to develop four hypotheses that will be the object of empirical verification in this study.

Hypothesis 1.

Exchange rate volatility directly hinders economic growth in emerging countries.

Hypothesis 2.

The emerging countries indirectly pay the cost of exchange rate volatility in terms of growth through one of its determinants.

Hypothesis 3.

Economic growth in emerging countries is penalized by both the direct and indirect effects of exchange rate volatility.

Hypothesis 4.

These effects differ between countries based on the chosen exchange rate regime. Consequently, we will establish the optimal exchange rate regime for emerging countries.

3. Methodology and Data

Despite the significance of empirical and theoretical studies to validate the intensity of the exchange rate’s impact on economic growth, this link has not been conclusively proven. Hence, this paper aims to explain how exchange rate volatility impacts economic growth under the different exchange rate regimes adopted by emerging countries.

Our analysis focuses on a panel of 14 emerging countries, including Morocco, Egypt, Turkey, Brazil, China, South Africa, Thailand, Tunisia, Nigeria, Indonesia, Mexico, Jordan, Peru, and Bolivia, from 1990 to 2020. Indeed, our survey of the economic literature has allowed us to draw empirical evidence on the direct impact of exchange rate volatility on economic growth. Additionally, there is a chance that exchange rate volatility may indirectly impact economic growth through its determinants, namely the degree of trade openness, inflation, the degree of financial system depth and development, external debt, and foreign indirect investment. The basic model of our study is presented in the following equation:

$$\begin{array}{c}{\mathrm{GDP}}_{\mathrm{it}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1 L{\mathrm{GDP}}_{\mathrm{it}}+{\mathsf{\alpha }}_2\mathrm{Exchange}\mathrm{rate}\mathrm{Volatility}{}_{\mathrm{it}}+{\mathsf{\alpha }}_3\mathrm{Degree}\mathrm{of}\mathrm{trade}\mathrm{openness}{}_{\mathrm{it}}+{\mathsf{\alpha }}_4\mathrm{Inflation}{}_{\mathrm{it}}\\ +{\mathsf{\alpha }}_5\mathrm{FDI}{}_{\mathrm{it}}+{\mathsf{\alpha }}_6\mathrm{Foreign}\mathrm{debt}{}_{\mathrm{it}}+{\mathsf{\alpha }}_7\mathrm{Financial}\mathrm{system}\mathrm{developpement}{}_{\mathrm{it}}+{\mathsf{\epsilon }}_{\mathrm{it}}\end{array}$$

(1)

The information on each of the variables used in our study is summarized in the

Table 1. A summary of variables used.

Variable	Measurements	Source	Observations
GDP	Real GDP per capita	World Bank database	Annual observations from 1990 to 2020
Exchange rate volatility	Real effective exchange rate (REER) volatility via GARCH Model	Bruegel database developed by Darvas (2021)	Monthly observations from 1990 to 2020
Degree of trade openness	(Exportations + importations)/GDP	World Bank database	Annual observations from 1990 to 2020
Inflation	Consumer price index (CPI)	World Bank database	Annual observations from 1990 to 2020
Foreign direct investment (FDI)	FDI/GDP	World Bank database	Annual observations from 1990 to 2020
Foreign debt	Total external debt/GNI	World Bank database	Annual observations from 1990 to 2020
Financial system development	M3/GDP	World Bank database	Annual observations from 1990 to 2020

Note that: For “Exchange rate volatility” variable, we will annualize it by calculating the mean of each year before its integration into the model.

below:

In the final stage of our research, included the exchange rate regime as a control variable to analyze the effects of exchange rate regime choice on this relationship. Exchange rate regime classification has historically generated a great deal of debate. Two different classifications can typically be distinguished. An institutional classification called “De jure” was derived from the reports published annually by the International Monetary Fund. However, several researchers have found discrepancies between the exchange rate regimes officially reported by countries to the IMF and those applied. To remedy this, the authors attempted to reclassify these regimes based on different criteria, giving rise to the second category of academic classification of exchange rate regimes called “De Facto.” These include, for example, Reinhart and Rogoff (2004), Levy-Yeyati and Sturzenegger (2005) classifications. For our study, we used the classification developed by Ilzetzki et al. (2017). These authors propose a database of exchange regime classifications (updated annually) for 194 countries from 1946 to 2020. It is based on the exchange rate volatility, measured on the informal parallel market (if it exists) and on the official exchange rate, while taking into account another critical parameter ‘‘the anchoring system”. Below (

Table 2. Classification of exchange rate regime.

Exchange Rate Regime	IRR Code
Fixed exchange rate regime (regime 1)	1,2,3,4
Intermediate exchange rate regime (regime 2)	5,6,7,8,9,10,11
Floating exchange rate regime (regime 3)	12,13,14,15

) we present the classification used:

We sorted the exchange rate regimes from the most rigid to the most flexible {1; 2; 3} = {Fixed; Intermediate; Float}. After including this variable, the model becomes:

$$\begin{array}{l}{\mathrm{GDP}}_{\mathrm{it}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1 L{\mathrm{GDP}}_{\mathrm{it}}+{\mathsf{\alpha }}_2{\mathrm{Exchange}\mathrm{rate}\mathrm{regime}}_{\mathrm{it}}+{\mathsf{\alpha }}_3{\mathrm{Exchange}\mathrm{rate}\mathrm{volatility}}_{\mathrm{it}}+{\mathsf{\alpha }}_4{\mathrm{Degree}\mathrm{of}\mathrm{trade}\mathrm{openness}}_{\mathrm{it}}\\ +{\mathsf{\alpha }}_5{\mathrm{Inflation}}_{\mathrm{it}}+{\mathsf{\alpha }}_6{\mathrm{FDI}}_{\mathrm{it}}+{\mathsf{\alpha }}_7{\mathrm{Foreign}\mathrm{Debt}}_{\mathrm{it}}+{\mathsf{\alpha }}_8{\mathrm{Financial}\mathrm{system}\mathrm{developpement}}_{\mathrm{it}}+{\mathsf{\epsilon }}_{\mathrm{it}}\end{array}$$

(2)

In our study, used the CS-ARDL panel model while taking into account the specifics of our data (as we will see later). This model is based on the panel auto-regressive distributed lags (Panel-ARDL) approach proposed by Pesaran and Smith (1995) and developed by Pesaran et al. (1999). This model has at least two particularities: First, it allows us to distinguish the short-run effect from the long-run effect. Second, it can produce inconsistent estimates of the average value of the parameters unless the coefficients are identical across countries. The general panel ARDL (p,q) model, according to Pesaran and Shin (1996), takes the following form:

$$Y_{it}={\omega }_i+{{\displaystyle \sum }}_{j=1}^p{б}_{i,j}Y{}_{i,t-j}+{{\displaystyle \sum }}_{j=0}^q{\mathsf{\phi }}_{i,j}X{}_{i,t-j}+{\mathsf{\epsilon }}_{it}$$

(3)

$i$ stands for the country with $i=1,2,3,\dots ,N.$ $t$ represents the time dimension. $j$ is the number of time lags. $X_{it}$ is the vector of explanatory variables. ${\omega }_i$ is the specific fixed effect of the individuals. $p$ and $q$ are lag orders, and ${\mathsf{\epsilon }}_{it}$ is the disturbing component.

The following Equation (4) is reparametrized as follows to take into account the adjustment coefficient and long-run dynamics:

$$\mathsf{\Delta }{Y}_{it}={\omega }_i+{\Omega }_i(Y{}_{i,t-1}-{\mathsf{\theta }}_{i}X{}_{i,t})+{{\displaystyle \sum }}_{j=1}^{p-1}{{б}^{\prime }}_{i,j}\mathsf{\Delta }Y{}_{i,t-j}+{{\displaystyle \sum }}_{j=0}^{q-1}{{\mathsf{\phi }}^{\prime }}_{i,j}\mathsf{\Delta }X{}_{i,t-j}+{\mathsf{\epsilon }}_{it}$$

(4)

where ${\mathsf{\theta }}_i$ indicates the long-run equilibrium relationship between $Y_{it}$ and $X_{it}$. ${б}^{\prime }$ and ${\mathsf{\phi }}^{\prime }$ represent the short-run coefficients linked with its past values and the variables of interest $X_{it}$. However, the sign of ${\Omega }_i$ indicates whether the variables have a long-term association. In addition, it presents the speed of adjustment while correcting for errors.

Three methods are used to estimate the dynamic ARDL panel data model: the mean group (MG), the pooled mean group estimator (PMG), and the dynamic fixed effect estimator (DFE). Indeed, one of the limitations faced by panel-ARDL estimators is the cross-sectional dependence. The cross-sectional independence, which is one of the conditions for using the Panel-ARDL model, remains a relatively strong assumption that may not be true when the sample observations are countries.

Chudik and Pesaran (2015) suggest applying Pesaran’s (2006) common correlated effects (CCE) approach in the context of panel ARDL models (CS-ARDL) to address the issue of error cross-sectional dependence. Thus, by including $Z$ = [$\sqrt[3]{T}$] lags of the cross-sectional averages, the CCE estimator gains consistency. Moreover, to account for common factors, this CS-ARDL approach appropriately adds additional lags of cross-sectional averages to the individual ARDL regressions. The CS-ARDL is displayed as follows:

$$Y_{it}={\omega }_i+{{\displaystyle \sum }}_{j=1}^p{б}_{i,j}Y{}_{i,t-j}+{{\displaystyle \sum }}_{j=0}^q{\mathsf{\phi }}_{i,j}X{}_{i,t-j}+{{\displaystyle \sum }}_{j=0}^z{\mathsf{{\rm Y}}}^{\prime }{}_{i,j}{\overline{\beta }}_{t-j}+{\mathsf{\epsilon }}_{it}$$

(5)

with ${\overline{\beta }}_{t-j}$ $=$(${\overline{y}}_{i,t-j}$, ${\overline{x}}_{i,t-j}$) given that: $\overline{y_t}$ = ${{\displaystyle \sum }}_{i=1}^N\frac{y_{i,t}}{N}$ and $\overline{x_t}$ = ${{\displaystyle \sum }}_{i=1}^N\frac{x_{i,t}}{N}$, and $Z$ being the number of lags of the cross-sectional averages to be included. We can calculate the mean group estimates in the CS-ARDL approach as follows:

$${\hat{\mathsf{\Phi }}}_{CS-ARDL,i}=\frac{{{\displaystyle \sum }}_{j=0}^q{\hat{\mathsf{\phi }}}_{i,j}}{{{\displaystyle \sum }}_{j=1}^p{\widehat{б}}_{i,j}}$$

(6)

Concerning modeling the volatility of the exchange rates in the countries on our panel, this analysis required the collection of historical monthly real effective exchange rates (REER) from the Bruegel database developed by Darvas (2021). In contrast, these authors suggest a variety of REER databases based on the number of trading partners. For our purposes, we have restricted our analysis to REER, considering 51 principal trading partners between January 1990 and December 2020, i.e., 364 observations.

To measure the volatility of our real effective exchange rates (REER), we calculated daily returns using the Napierian logarithm of the monthly rates of each rate using the following method:

$$REER Retur{n}_t=Log\left(\frac{REER{}_t}{REER{}_{t-1}}\right)$$

(7)

Various measures have been used to capture exchange rate volatility in the literature. The first measures are based on historical volatility and focus on dispersion indicators like the standard deviation and coefficient of variation (Kenen and Rodrik 1986; Dell’Ariccia 1999) or the long-term exchange rate uncertainty proposed by Perée and Steinherr (1989). However, historical volatility does not consider exchange rate uncertainty, representing the unanticipated part of exchange rate fluctuations. Then, it seemed more evident to use the concept of conditional volatility as measured by the GARCH (p,q) models family introduced by Engle (1982) and generalized by several authors, including Bollerslev (1986) and Gouriéroux (1997). These models have shown their performance in measuring the volatility of exchange rates (see Syarifuddin et al. 2014; Barguellil et al. 2018) as well as the volatility of stock market indexes (see Ameziane and Benyacoub 2022). We can write the general form of the GARCH (p,q) model as follows:

$${\sigma }_t{}^2=ɲ+{{\displaystyle \sum }}_{i=1}^P{\beta }_i\sigma {}^2{}_{t-1}+{{\displaystyle \sum }}_{j=1}^Q{\alpha }_j\epsilon {}^2{}_{t-j}$$

(8)

where $ɲ$ is the long-run volatility with: $ɲ>0$; ${\beta }_i\ge 0;i=1,\dots ,p$ and ${\alpha }_j>0$; j $=1,\dots ,q$.

The GARCH model is a generalized autoregressive model that captures the volatility clusters of returns through conditional variance. However, as shown in the model function (8), the GARCH model can generate two crucial parameters: the unconditional variance, which is constant and more concerned with the long-term behavior of the time series. The second is the conditional variance, which is a one-period ahead estimate of the variance calculated on any past information thought to be relevant, and which converges in the long run to the value of the first. In order to ensure that the exchange rate returns’ unconditional variance is constant, many researchers frequently presume that a stable GARCH process models the volatility information in the conditional variance. However, financial indicators are always subject to sudden shocks that cause abrupt changes in the unconditional variance of their returns, thereby undermining their long-term stagnation. To remedy this situation, model validation tests are required after determining the appropriate GARCH model to ensure that the conditional variance summarizes all volatility information.

4. Empirical Analysis

We have split this research phase into three sections: First, we present the different steps of REER volatility modeling using the GARCH (1,1). In the second part, we analyze the different specification conditions of the CS-ARDL panel model. A third part is devoted to presenting the estimation findings according to the selected model.

4.1. GARCH Model Specification

4.1.1. Normality and Stationarity Analysis

The first phase of our model’s development is devoted to analyzing two preconditions: the normality and stationarity of the time series. The descriptive statistics listed below form the basis of the normality study (

Table 3. Descriptive statistics of the REER return series.

		Mean	Median	Std Dev	Skewness	Kurtosis	Jarque–Bera
REER return	Morocco	0.000312	−0.000237	0.008407	0.447027	12.54703	1425.15 ***
	Egypt	0.003264	0.004749	0.021498	−0.467962	7.125809	277.4230 ***
	Turkey	−0.000486	0.003296	0.036726	−1.537170	11.20015	1188.758 ***
	Brazil	−0.001072	0.000124	0.041902	−0.886052	7.807281	406.8797 ***
	China	0.001162	0.001925	0.015185	−2.176065	16.64212	3178.250 ***
	South Africa	−0.000930	0.000582	0.030645	−0.540152	6.156610	172.5343 ***
	Thailand	0.00326	0.001560	0.020634	−1.293796	26.15978	8417.601 ***
	Tunisia	−0.000909	−0.000544	0.009571	−0.411172	8.354141	454.8176 ***
	Nigeria	0.004832	0.007129	0.039589	−4.194360	44.00281	27149.81 ***
	Indonesia	−0.000210	0.000742	0.053708	−3.735562	50.14433	35315.28 ***
	Mexico	$-7.15\times {10}^{-5}$	0.003837	0.122708	−3.247965	29.85828	11835.25 ***
	Jordan	0.000946	0.000367	0.014151	0.517037	4.209452	39.24729 ***
	Peru	−0.000414	0.001146	0.035408	−5.040494	56.22029	45477.93 ***
	Bolivia	0.000752	0.000383	0.011745	0.256123	3.515384	8.184254 ***

Note: *** indicates the significance level of 1%.

).

Our distributions’ non-normality is confirmed using descriptive statistics. On the one hand, the absolute values of the skewness coefficients deviate from 2. On the other hand, the kurtosis coefficient for our distributions is different from seven (Kim 2013)3. The Jarque–Bera significant coefficients for all rate series’ support these findings (the hypothesis of the non-normality of the distributions).

The Dickey and Fuller (1979) test is the stationarity test applied in our research. Its objective is to test the null hypothesis of no unit root against the alternative hypothesis of the presence of a unit root.

The ADF test results (

Table 4. Dickey–Fuller Test results.

		Trend and Intercept		Intercept		None
		T-Stat	Prob.	T-Stat	Prob.	T-Stat	Prob.
REER Return	Morocco	−21.78205	0.0000	−21.74081	0.0000	−21.73509	0.0000
	Egypt	−13.39497	0.0000	−13.40213	0.0000	−9.360131	0.0000
	Turkey	−13.00595	0.0000	−12.95937	0.0000	−12.97487	0.0000
	Brazil	−13.74271	0.0000	−13.75004	0.0000	−13.76528	0.0000
	China	−16.80247	0.0000	−16.84384	0.0000	−16.71348	0.0000
	South Africa	−15.62128	0.0000	−15.64262	0.0000	−15.65239	0.0000
	Thailand	−14.07934	0.0000	−14.07730	0.0000	−14.09324	0.0000
	Tunisia	−14.56372	0.0000	−14.54470	0.0000	−14.46113	0.0000
	Nigeria	−13.55185	0.0000	−13.46698	0.0000	−13.23501	0.0000
	Indonesia	−16.48059	0.0000	−16.49835	0.0000	−16.52051	0.0000
	Mexico	−15.1429	0.0000	−15.14028	0.0000	−15.16076	0.0000
	Jordan	−14.34529	0.0000	−14.36668	0.0000	−14.33124	0.0000
	Peru	−13.37768	0.0000	−13.37742	0.0000	−13.39030	0.0000
	Bolivia	−12.33671	0.0000	−12.25317	0.0000	−12.22659	0.0000

) reveals the absence of a unit root in our time series. As a result, the processes related to our series are consistent with the level stationarity property. Consequently, our initial two requirements for elaborating a GARCH model, non-normality and stationarity, have been successfully established.

4.1.2. ARMA Analysis Process

The second crucial step in the process of GARCH model determination is determining the ARMA models. It combines the autoregressive part AR, constituted by a finite linear combination of the past values of the process, with a moving average part MA formed by a limited linear combination of the past values of white noise. These are, therefore, models of order (p,q) whose process is generated by a combination of past values and past errors. Process determination ARMA orders (p,q) are based on simple and partial autocorrelation correlograms. The correlograms of the returns reveal insignificant autocorrelations even for high lags. Thus, there are a multitude of ARMA models (p,q) to use, of which the models ARMA (1,1) and ARMA (1,0) are likely.

4.1.3. ARCH Effect Analysis

After identifying the ARMA (p,q) process, we move on to the third phase of our research to look into the possibility of the ARCH effects’ existence. In general, ARCH (autoregressive conditional heteroscedasticity) models allow for the estimation of instantaneous chronic volatilities that depend on the past and for dynamic forecasts in terms of means and variances (Engle 1982).

The results of the ARCH heteroskedasticity test (

Table 5. The ARCH heteroskedasticity test results.

		Heteroskedasticity Test: ARCH
		F-Statistic	Prob. F	Obs Rsquared	Prob. Chi Square
REER return	Morocco	17.29655	0.0000	16.61164	0.0000
	Egypt	6.922622	0.0089	6.831971	0.0090
	Turkey	13.03979	0.0003	12.66298	0.0004
	Brazil	58.22206	0.0000	50.56009	0.0000
	China	6.557652	0.0108	6.478070	0.0109
	South Africa	3.915363	0.0486	3.895253	0.0484
	Thailand	46.35219	0.0000	41.40260	0.0000
	Tunisia	29.32515	0.0000	27.31344	0.0000
	Nigeria	5.442191	0.0202	5.392162	0.0202
	Indonesia	99.44115	0.0000	78.75625	0.0000
	Mexico	26.86426	0.0000	26.17691	0.0000
	Jordan	4.000524	0.0462	3.979068	0.0461
	Peru	19.92205	0.0000	19.00401	0.0000
	Bolivia	4.683871	0.0311	4.650231	0.0310

) are significant, confirming the null hypothesis of the existence of the ARCH effect in our time series. As a result, we can use the ARCH (1) and GARCH (1,1) models to simulate the REER return’s volatility. The GARCH model, however, is preferred because it is more accurate.

To validate our model, we analyze the normality and the absence of the ARCH effect in the series of squared residuals. We present the results in the following table (

Table 6. Validation tests of the GARCH model.

Validation Tests of the GARCH (1.1) Model
		Q (20)	Q2 (20)	ARCH 1-10
REER Return	Morocco	18.3329 (0.4339347)	11.3275 (0.8799475)	0.90473 (0.5295)
	Egypt	15.4577 (0.6303388)	14.2312 (0.7139015)	1.2186 (0.2775)
	Turkey	19.5698 (0.3575539)	5.49497 (0.9978522)	0.28138 (0.9848)
	Brazil	23.3262 (0.1783642)	13.6104 (0.7541156)	0.43752 (0.9270)
	China	10.0384 (0.2623423)	16.8055 (0.5365104)	1.2018 (0.2884)
	South Africa	19.6164 (0.4179836)	19.7298 (0.3482343)	1.1029 (0.3589)
	Thailand	16.9175 (0.5287857)	17.6937 (0.4759965)	1.0546 (0.3987)
	Tunisia	21.1785 (0.3270212)	23.5922 (0.1688471)	1.3031 (0.2271)
	Nigeria	18.9419 (0.3954077)	20.2343 (0.3197841)	0.30567 (0.9792)
	Indonesia	23.7845 (0.1622139)	12.1431 (0.8397614)	0.59759 (0.8151)
	Mexico	19.2688 (0.3754485)	8.37274 (0.9725557)	0.32059 (0.9755)
	Jordan	19.5176 (0.3606243)	10.8110 (0.9022039)	0.61233 (0.8027)
	Peru	19.5529 (0.1622139)	9.01299 (0.9594410)	0.75130 (0.6758)
	Bolivia	21.1850 (0.2701801)	12.9625 (0.7937978)	0.64689 (0.7732)

):

The findings (Table 6) indicate that all parameters deviate significantly from zero. The models support the tests for the absence of residual and squared residual autocorrelation. Additionally, it supports the ARCH effect’s absence (up to order 10).

As a result, we proved that the GARCH (1,1) model could accurately estimate REER volatility. In other words, the conditional variance generated by the GARCH (1,1) model models all the information related to the volatility of our time series. Hence, there is the possibility of using it as a proxy for the volatility of exchange rates.

4.2. CS-ARDL Panel Model Specification

The development and specification of the CS-ARDL panel model require the verification of preliminary conditions:

-

The order of integration of the variables.

-

The homogeneity of the panel.

-

Cross-sectional dependency.

-

The existence of a long-term link between the variables, notably the presence of cointegration.

4.2.1. Cross-Sectional Dependency Test

We use the test developed by Pesaran (2015) to test the existence of cross-sectional dependency. According to Kapetanios et al. (2011), the assumption of cross-sectional independence of the error is fundamental to ensuring the convergence of estimators obtained by the group mean method. Ignoring this dependence could result in biased estimated parameters and incorrect standard deviation calculations. The results are as follows (

Table 7. Cross-sectional dependence test.

CD-Test for Cross-Sectional Dependence
Variables	CD-Test	p-Value	Average Joint T
Log GDP	48.763	0.000	31.00
Log REER volatility	3.291	0.001	31.00
Log degree of openness	17.985	0.000	31.00
Log CPI	48.77	0.000	31.00
Log M3/GDP	22.18	0.000	31.00
Log foreign debt	17.28	0.000	31.00
Log FDI/GDP	11.207	0.000	31.00

):

For all the variables, we achieved significant results at 1%. As a result, our panel is characterized by cross-sectional dependency.

4.2.2. Homogeneity Test

To verify this condition, we will opt for two complementary tests. The first is the Swamy (1970) test, which considered the dispersion of individual slope estimates from an appropriate pooled estimator as a basis for slope homogeneity. The second is the Pesaran and Yamagata (2008) test, which compares the weighted difference between the cross-sectional unit-specific estimate and a weighted pooled estimate. The results of both tests are presented in

Table 8. Slope heterogeneous test.

Swamy’s Test		Pesaran and Yamagata’s Test
Chi-2 (91)	33,126.27 (0.000)	Delta	14.745 (0.000)
		Delta adj.	17.118 (0.000)

:

The Prob > Chi-2 Swamy test (1970), in addition to the p-values of the Pesaran and Yamagata test (2008), are significant at the 1% significance level. Hence the rejection of the null hypothesis and the confirmation of the heterogeneity of the data-generating process (slope heterogeneity existed).

4.2.3. Unit Root Test

In our study, we use two famous tests: the CIPS test for unit roots in heterogeneous panels, developed by Pesaran (2007), and pescadf test, developed by Pesaran (2003), which runs the t-test by taking into consideration the issues of heterogeneity and cross-sectional dependence. These tests allow us to test the null hypothesis of the existence of unit roots in the panels against the alternative hypothesis indicating their absence (stationary variables). The following table (

Table 9. Unit Root Test results.

	CIPS Test		CADF Test
Variables	Level	First Difference	Level	First Difference
Log GDP	−1.898 (−2.44)	−3.559 *** (−2.45)	−1.898 (0.306)	−3.559 (0.000)
Log REER volatility	−3.643 *** (−2.44)	−	−3.643 (0.000)	−
Log degree of openness	−1.436 (−2.44)	−4.506 *** (−2.45)	−1.436 (0.908)	−4.506 (0.000)
Log CPI	−1.872 (−2.44)	−4.307 *** (−2.45)	−1.872 (0.342)	−4.307 (0.000)
Log M3/GDP	−2.235 ** (−2.25)	−	−2.235 (0.032)	−
Log foreign debt	−2.905 *** (−2.44)	−	−2.905 *** (0.000)	−
Log FDI/GDP	−3.110 *** (−2.44)	−	−3.110 *** (0.000)	−

Note: *** and ** indicate the significance levels of 1% and 5%, respectively.

) shows the results of the test:

Our findings from both tests were consistent. The REER volatility (Log REER volatility), the degree of depth and development of the financial system (Log M3/GDP), the volume of foreign debt (Log foreign debt), and foreign direct investment (Log FDI/GDP) are stationary variables at the level. In contrast, the variables of economic growth (Log GDP), the degree of trade openness (Log degree of openness), and inflation (Log IPC) are integrated into order one I(1). Therefore, our results verified the mixture condition of integration orders I(0) and I(1).

4.2.4. Cointegration Test

Verifying the existence of a long-term relationship between the model’s variables, particularly the cointegration, is one of the requirements for using a panel CS-ARDL model. Nevertheless, during this research phase, we also had to consider the cointegrating variables’ heterogeneity and cross-sectional dependence. Consequently, we use the Westerlund (2007) cointegration test using error correction. The results of the test are shown below in

Table 10. Westerlund ECM panel cointegration tests.

Statistics	Value	Z-Value	Robust p-Value
Gt	−0.199	8.169	0.820
Ga	−0.646	5.926	0.030
Pt	−0.203	6.493	0.020
Pa	−0.055	4.253	0.040

:

Our results show the significance of three out of four Westerlund statistics at the 5% level. It confirms the presence of a long-term relationship between our variables.

In the presence of a mix of degrees of integration of our variables (I(0) and I(1)), the presence of a cross-sectional dependency, heterogeneity slope, and the existence of a long-run cointegration between the variables, the adequate model to adopt in our study is the Panel CS-ARDL model. The last step of our modeling process is to determine the optimal lags. Based on the AIC information criterion, the model to be used for our study is CS-PARDL (1 0 0 0 1 0 0).

5. Results and Discussion

This part of our paper is devoted to a series of empirical evaluations with a dual purpose. The first one is to analyze exchange rate volatility’s direct or indirect impact on economic growth. The second one is to examine whether the exchange rate regime adopted is an effective tool to reduce this effect. We will base our analysis on two main tools to achieve our goal. The results of our model, the CS-ARDL panel model (

Table 11. Panel CS-ARDL estimation results.

Common Correlated Effects Estimator–(CS-ARDL Panel)
Variables	Coefficients	Std. Err	p-Value
Short-run estimation
L.Log GDP	0.5158881	0.0809883	0.000 ***
Log REER volatility	−0.0134248	0.0073008	0.066 *
Log degree of openness	−0.0088341	0.0197377	0.654
Log CPI	−0.1736491	0.081663	0.033 **
Log FDI/GDP	0.0090205	0.0033645	0.007 ***
Log foreign debt	−0.0170404	0.008012	0.033 **
L.Log M3/GDP	0.092221	0.0252932	0.000 ***
Long-run estimation
Log REER volatility	−0.0751851	0.0383397	0.050 **
Log degree of openness	0.020038	0.0652003	0.759
Log CPI	−0.3716829	0.1441866	0.010 ***
Log FDI/GDP	0.0307217	0.0121971	0.012 **
Log foreign debt	−0.0308979	0.0228938	0.177
Log M3/GDP	0.2546056	0.1137025	0.025 **

Note: ***, **, and * indicate the significance levels of 1%, 5%, and 10%, respectively.

), and the Granger non-causality test created by Dumitrescu and Hurlin (2012) in

Table 12. Granger non-causality test results.

Dumitrescu and Hurlin (2012) Granger Non-Causality Test
	W-Bar	Z-Bar	Z-Bar Tilde
	Log REER Volatiliy
Log CPI	3.6245	6.9438 *** (0.0000)	5.8628 *** (0.0000)
Log FDI/GDP	1.8573	2.2682 ** (0.0233)	1.7910 * (0.0733)
Log foreign debt	2.5267	4.0392 *** (0.0001)	3.3333 *** (0.0009)
Log M3/GDP	2.1524	3.0490 *** (0.0023)	2.4710 ** (0.0135)

Note: ***, **, and * indicate the significance levels of 1%, 5%, and 10%, respectively.

.

The current instability of the global economic environment, reflected in low growth rates, has led us to examine the determinants of growth in emerging countries. According to the CS-ARDL model’s estimation, all economic growth determinants studied in emerging countries, except the degree of trade openness, appear to affect it in the short term. The impact is positive for the volume of FDI and the depth of the financial system. In contrast, the level of external debt and inflation harms economic growth. However, the CS-ARDL long-run estimation shows that, except for the degree of openness and the level of external debt, whose effects are felt only in the short term, the other variables determining economic growth (inflation, volume of FDI, and depth and development of the financial system) have maintained their effect even in the long term.

Regarding the crucial component of our model, exchange rate volatility, we observe that it has a significant negative short-term impact at the 10% level. Economic growth is penalized by almost 0.013% for every 1% increase in REER volatility. However, our results show that this effect worsens in the long run, with a 0.075% loss in economic growth for every 1% increase in exchange rate volatility.

These results allow us to conclude that the variations in growth return to risk factors (related to the direct impact of exchange rate volatility on economic growth), liquidity factors (related to inflation as a monetary phenomenon), as well as other micro- and macroeconomic factors such as (the level of FDI, the level of development and depth of the financial system, and the level of external debt).

To scrutinize the indirect effect of exchange rate volatility on economic growth, we must examine the impact of exchange rate volatility on the determinants of GDP. We accomplish this purpose using the Granger non-causality test created by Dumitrescu and Hurlin (2012). We point out that causality analysis is carried out with the determinants that have shown significant results in the CS-ARDL model analysis, notably inflation (log CPI), the level of foreign direct investment (log FDI/GDP), the level of foreign debt (log foreign debt), and the level of development and depth of the financial system (log M3/GDP). The following are the results (Table 12):

According to the findings, all determinants can commonly or singularly act as a transmission channel for the effect of exchange rate volatility on economic growth.

Furthermore, REER volatility penalizes long-term and short-term economic growth in emerging countries. Nevertheless, this negative effect manifests in two aspects: a direct negative effect and an indirect negative impact transmitted via inflation (the pass-through phenomenon), the FDI, the level of external debt, and the level of development and depth of the financial system.

To link our research with the predetermined goals, we conduct the same analysis while integrating the adopted exchange rate regime as a control variable. This step will allow us to analyze the impact of exchange rate volatility on economic growth under each type of exchange rate regime. Below we present the estimation of our CS-ARDL model (

Table 13. Panel CS-ARDL estimation results by exchange rate regime.

Common Correlated Effects Estimator–(CS-ARDL)
Variables	Fixed ERR			Intermediate ERR			Floating ERR
	Coeff	Std. Err	p-Value	Coeff	Std. Err	p-Value	Coeff	Std. Err	p-Value
Short-run estimation
L.Log GDP	0.6182333	0.1716174	0.000 ***	0.8587536	0.0512364	0.000 ***	0.4402354	0.137329	0.001 ***
Log REER volatility	−0.0579011	0.0021231	0.000 ***	−0.0039421	0.0040626	0.332	−0.0171749	0.0094374	0.069 *
Log degree openness	0.3722057	0.0341857	0.000 ***	−0.1523318	0.0109061	0.000 ***	0.1006791	0.0413309	0.015 **
Log CPI	0.3420491	0.0602633	0.000 ***	−0.5102768	0.1253434	0.000 ***	−0.0841091	0.0312508	0.007 ***
Log FDI/GDP	0.0055867	0.0083584	0.504	0.0365185	0.0041678	0.000 ***	0.0063403	0.0045638	0.165
Log foreign debt	0.0213044	0.0117241	0.069 *	0.0482735	0.00937	0.000 ***	0.0037782	0.0189084	0.842
L.Log M3/GDP	0.3395888	0.0064759	0.000 ***	0.2555339	0.0722258	0.000 ***	−0.0298541	0.0497969	0.549
Long-run estimation
Log REER volatility	−0.1869441	0.0784766	0.017 ***	−0.0441529	0.0447785	0.324	−0.0231588	0.0113464	0.041 **
Log degree of openness	1.272323	0.6614994	0.054 *	−1.209643	0.3615779	0.001 ***	0.1931079	0.0726293	0.008 ***
Log CPI	1.211809	0.7026037	0.085 *	−4.530748	2.530914	0.073 *	−0.2583357	0.1786905	0.148
Log FDI/GDP	0.006005	0.0191947	0.754	0.310045	0.1419746	0.029 **	0.0150599	0.0074	0.042 **
Log foreign debt	0.0526364	0.0070482	0.000 ***	0.4212627	0.2191484	0.055 *	−0.0305881	0.0521901	0.558
Log M3/GDP	1.124356	0.5224001	0.031 **	2.296852	1.344517	0.088 *	−0.0753632	0.1188511	0.526

Note: ***, **, and * indicate the significance levels of 1%, 5%, and 10%, respectively.

):

Regarding the Dumitrescu and Hurlin (2012) Granger non-causality test results for each exchange rate regime, They are presented in the

Table 14. Granger non-causality test results by exchange rate regime.

	Dumitrescu and Hurlin (2012) Granger Non-Causality Test
	Fixed ERR			Intermediate ERR			Floating ERR
	Log REER Volatiliy
	W-Bar	Z-Bar	Z-Bar Tilde	W-Bar	Z-Bar	Z-Bar Tilde	W-Bar	Z-Bar	Z-Bar Tilde
Log degree of openness	1.5106	1.0212 (0.3072)	0.7500 (0.4533)	1.3476	0.8863 (0.3754)	0.5943 (0.5523)	1.1000	0.2122 (0.8319)	0.0370 (0.9705)
Log CPI	4.2536	6.5072 (0.0000)	5.5276 (0.0000)	3.4688	6.2941 (0.0000)	5.3037 (0.0000)	5.1013	8.7003 (0.0000)	7.4290 (0.0000)
Log FDI/GDP	-	-	-	1.6630	1.6902 (0.0910)	1.2943 (0.1956)	2.5985	3.3908 (0.0007)	2.8052 (0.0050)
Log foreign debt	3.9612	5.9223 (0.0000)	5.0182 (0.0000)	2.6410	4.1838 (0.0000)	3.4659 (0.0005)	-	-	-
Log M3/GDP	0.6316	−0.7368 (0.4613)	−0.7810 (0.4348)	1.8304	2.1172 (0.0342)	1.6662 (0.0957)	-	-	-

below:

By segmenting our panel according to the exchange rate regime, we concluded that only countries that adopt fixed and floating exchange rate regimes suffer from the direct impact of exchange rate volatility on economic growth. For the first set of countries, increases in the REER volatility by 1% degrade economic growth by 0.058% in the short run and by 0.187% in the long run. However, for countries that permit their currency to float freely, these rates are 0.017% and 0.023%, respectively (see coefficients in Table 10). If we compare these two regimes, we can see that in the event of excessive exchange rate volatility, emerging countries adopting a fixed exchange rate regime suffer more in terms of growth than those that let their currency float freely. In contrast, intermediate exchange rate regimes protect against the direct effects of exchange rate volatility on economic growth.

Nevertheless, exchange rate volatility can harm GDP through the channel of economic growth determinants (the determinants that showed significant results in Table 10) regardless of the adopted exchange rate regime. For countries with fixed and intermediate regimes, the impact can be transmitted through the inflation rate and/or the level of external debt. However, for countries liberalizing their exchange rate regime, it can be mediated through inflation and/or the level of FDI (see Table 11).

6. Conclusions

Economic growth is one of the most exciting areas to study for many reasons. Despite the great relevance and importance of the issues involved, this one still presents many unknowns and major challenges to discuss. For this reason, the objective of our document is dual. On the one hand, we analyze exchange rate volatility’s direct and indirect effects on economic growth. On the other hand, we examine the role of the exchange rate regime adopted in mitigating this impact.

Our econometric modeling is organized as follows: In the first part, we used the GARCH (1,1) model to generate the conditional variance of the exchange rates in order to use it later as a proxy variable for the exchange rate volatility. The second part of our study, which provided rich answers to our questions, was carried out using the CS-ARDL panel model. The empirical findings can be summarized into three main points. First, exchange rate volatility penalizes emerging countries’ economic growth directly and indirectly (confirmation of Hypothesis 3). Second, by examining our panel according to the exchange rate regime adopted, we found that countries adopting bipolar exchange rate regimes, notably fixed and floating exchange rate regimes, are the most predisposed to the direct and indirect effects of exchange rate shocks. However, since the intermediate regimes are not immune, they at least provide a shield protecting countries from the direct impact of exchange rate volatility (confirmation of Hypothesis 4). Our results align with the results in this framework by Ghosh and Ostry (2009) by asking the question: Should countries fix, float, or choose something in between? This relative underperformance of intermediate exchange regimes is due to their combination of flexibility and rigidity or “happy balance between pegs and free floats” as Ghosh and Ostry qualified it. On the one hand, this regime category provides a tolerable level of exchange rate volatility. On the other hand, beyond this level, the country benefits from the second characteristic, which is the possibility of intervention to reduce economic vulnerability.