The Impact of Interest Rate on Exchange Rate Within ASEAN Countries: Evidence from Linear and Nonlinear ARDL Frameworks

Abstract

This study examines the symmetric and/or asymmetric effects of changes in the interest rate on exchange rate of the ASEAN countries. It further aims to compare these linkages by using a dataset consisting of 48–68 quarterly data items, ranging over the period 2002–2017, of the ASEAN countries. Using both the linear autoregressive distributed lag (ARDL) and nonlinear ARDL (NARDL) approaches, the findings indicate that these effects vary from one country to another. We observe that changes in interest rates have short-run symmetric effects on the exchange rates, which also hold in the long run for five ASEAN countries, namely, Cambodia, Malaysia, Thailand, Vietnam, and Singapore. On the other hand, changes in interest rates have asymmetric (negative) effects on the exchange rates, which also hold in the long run for seven ASEAN countries, namely, Cambodia, Indonesia, Malaysia, the Philippines, Singapore, Thailand, and Vietnam.

Keywords

Exchange rate interest rate linear ARDL NARDL ASEAN

Introduction

Various studies on the linkage between the interest rate and exchange rate have been conducted to understand their relationship. One of the economic models, known as the Mundell–Fleming model, proposes that in the short run, a decrease in the domestic country’s interest rates encourages capital outflow, which causes a deficiency in the balance of payments. One way of tackling this situation is by depreciating the home country’s currency through increasing net exports. According to several studies (Branson, 1981; Branson & Halttunen, 1979; Branson et al., 1977), when an increase in the domestic interest rate occurs, interest-bearing assets, such as coupons, become more profitable for lenders, driving the ownership of more assets for lenders. Thus, attracting more foreign investment can lead to an appreciation of the currency due to the rise in exchange rate. In fact, exchange rate can move in either direction, and the interest rate can be used as a controlling and monitoring tool by the governments.

The ASEAN countries have been facing notable economic vulnerabilities, such as foreign exchange volatility and unstandardized foreign exchange rates, among them. These economic vulnerabilities mainly occur due to the different responses from ASEAN countries to the external ups and downs that affect the management and fluctuations of ASEAN countries’ exchange rates. Thailand Development Research Institute’s President Chalongphob Sussangkarn claimed, “Recently when the United States revised its interest rate, the responses from (ASEAN) countries were not the same” (The Star, ASEAN Need A Mechanism, 2017). This lack of coordination leads to costlier cross-border transactions and acts as a discouragement for the steps taken for ASEAN to become a large economic bloc. The former Malaysian Second Minister of International Trade and Industry Ong Ka Chuan stated, “Some economists have forecasted the emergence of the ASEAN (AEC) economy as the fourth largest economic bloc after the European Union (E.U.), China and India by 2030 due to the continuous growth of the overall ASEAN GDP” (Sin, 2016). However, this growth will remain slow and fragile against external pressures, such as an increase in the US interest rate, especially if the lack of coordination among the ASEAN members continues. Therefore, the ASEAN countries should start integrating their efforts to cushion such external pressures by having common standards and responding in similar ways.

The ASEAN countries have to create an agreed-upon method for dealing with foreign exchange volatility and the unstandardized foreign exchange rates among themselves. One of the common standards that the central banks in ASEAN should adopt is similar monetary policies in terms of interest rate intervention and the range of inflation sustainability. However, in this study, the focus is on understanding the importance of the linkage between the interest rates and exchange rates among ASEAN economies.

Several practitioners have studied the relationship between interest rate and exchange rate using different methods. Hacker et al. (2014) used wavelet analysis and found that there was significant empirical evidence that nominal interest rate differential is driving the exchange rate volatility, specifically, more positive linkages in the long run and more negative relationships in the short run. Lily et al. (2011) found that there is a significant effect of uncovered interest rate parity upon exchange rate fluctuations, since the findings of autoregressive conditional heteroscedasticity and generalized autoregressive conditional heteroscedasticity (GARCH) models were significant in the GARCH model. Hnatkovska et al. (2013) have mathematically proven that there is a non-monotonic relationship between long-term interest rates and the exchange rate, by examining a data set of 80 different countries in different regions. Sollis and Wohar (2006) used symmetric and asymmetric threshold cointegration tests to expose evidence of long-run nonlinearity in the linkage of real exchange rate and real interest rate.

On the other hand, Abbas et al. (2012), using regression analysis, found that there is a nonsignificant effect of interest rate upon the exchange rate. Saraç and Karagöz (2016) found no relationship of the effects of a higher interest rate with depreciation of exchange rate, by using the frequency domain Granger causality test. From these previously mentioned studies, we can conclude that there is a mix of both symmetric and asymmetric empirical evidence for the relationship between interest rate and exchange rate.

The relationship between interest rate and exchange rate remains prominent; however, the deficiency of clear empirical evidence of the effect of interest rate on exchange rate without exception is more ambivalent from the perception of an economist. Usually, the classic policy action that is taken by central banks and policymakers is to use the short-term interest rate in order to manipulate currency values, as well as market monetary conditions in general terms. However, in case we assume that there is no linkage at all in the data, thereupon what is the reason that keeps policymakers from utilizing interest rate mechanisms in order to manipulate the exchange rate? Are some of the main aspects of the exchange rate–interest rate linkage getting camouflaged in the classic time series empirical findings? This study aims to uncover the linkages between interest rate and exchange rate using the new, and more precise, nonlinear autoregressive distributed lag (NARDL) method to help policymakers in the ASEAN countries create similar outcomes in their respective economies.

Literature Review

Several authors have used a direct-causality method to investigate the causality between the exchange rate and interest rate. They have found no significant causality flowing from the interest rate to exchange rate (Gül et al., 2007; Kayhan et al., 2013; Tari & Abasiz, 2009). Kisaka et al. (2014) used the Granger causality test to investigate the long-horizon and short-horizon correlation between the exchange rate and interest rate in Kenya. Their findings exhibit an existing connection between the interest rate and exchange rate in the long horizon. A flow of causality from the interest rate to the exchange rate is found as evidence of a unidirectional causality in Kenya. Therefore, the authors recommend that the Kenyan financial sector utilize the interest rate as a way to smooth the foreign exchange rate over the long run. Another study (Saraç & Karagöz, 2016) tried to shed light on the causal relationship between the interest and exchange rates primarily in the short run. However, the results exhibited no relationship between the two.

MacDonald and Nagayasu (2000) used cointegration analysis within a panel study to examine the long-horizon relationship between the real exchange and interest rates and found a very weak link in long-term linkage when having a constant equilibrium in the exchange rate. Asari et al. (2011) examined the linkage between the interest rate, inflation rate, and exchange rate in Malaysia over the period 1999–2009 using the vector error correction model (VECM) cointegration analysis technique. They found evidence that the inflation rate affects the interest rate and that the interest rate affects the volatility of the exchange rate in Malaysia. Akçağlayan (2008) studied the impact of interest rate intervention on foreign exchange rate utilizing an error correction model during the period of currency crisis in 2001 in Turkey and found evidence that a rise in the interest rate causes domestic currency to depreciate. Another study, by Gümüş (2002), used a VECM and weekly data to find out if increasing the interest rate leads to a long-run impact on exchange rate depreciation. According to Sollis and Wohar (2006), who used threshold cointegration to test nonlinearity among the exchange rate and interest rate, the equilibrium error’s nonlinear mean decline was caused by adjusting the real exchange rate in the nonlinear short horizon to destabilize the equilibrium. In the case of existing threshold cointegration, they found stronger evidence of mean decline whenever the equilibrium error was in the negative state, rather than when it was positive.

Hnatkovska et al. (2012) utilized the vector autoregressive (VAR) model to examine the relationship between the interest rate and exchange rate and mainly looked for a contemporary linkage. Their null hypothesis states that the interest rate has no long-horizon impacts on the real exchange rate. A number of monetary theory models have used this sort of standardized assumption of neutrality, such as Clarida and Gali (1994) and Bjørnland (2009). Another study, by Berument (2007), used the VAR model in small, open economies for the period 1986–2000. In his model, he considered the spread between the depreciation rate and interbank interest rate and found that it was involved with lower income and overall prices, and the appreciation of the domestic currency. However, when those variables were tested separately in the VAR analysis, a rise in the interbank rate indicated a permanent depreciation of domestic currency, a situation that has been referred to as the exchange rate puzzle. Also, Agénor et al. (1997) used the VAR method in their model for testing and found that shocks of the interest rate differential can result in a huge response from the exchange rate, whereby the latter responds to a shock in both a positive and a negative way. This finding has been used as supporting evidence for having tighter central bank intervention in Turkey.

Fitted GARCH models seem to show persistence and serial correlation in exchange rate variance (Engle & Bollerslev, 1986; Holland Domowitz & Hakkio, 1985). Therefore, the prediction of future volatility can be easily determined by an exchange rate’s current and previous volatility; however, differences in the trajectories may appear, especially in those that have matching initial conditions (Rasband, 1990). Several studies have adopted the GARCH model in their analysis, as described below. Huang et al. (2010) proposed that the linkage between the interest rate and exchange rate is based on time invariance, during their sample phases. Therefore, they utilized a time-varying parameter model with GARCH errors in their study on Korea, Thailand, and Indonesia to uncover the dynamic forces that contribute to this relationship. They found that there is no direct relationship between the interest rate and currency, with the exception of Korea, where the interest rate seems to influence the exchange rate. Chow and Kim (2006) used a bivariate GARCH model in order to seize the dynamic interaction between the fluctuations of the interest rate and exchange rate more efficiently, rather than having to model each variable exclusively using a single-equation GARCH model. According to Goyal and Arora (2012), time series analysis can appropriately determine the impact of several intervention variables on the exchange rate, because event analysis can only focus on 2–3 variables at a time and their findings are increasingly doubtable. Thus, in their study, the estimators of the exponential GARCH (EGARCH) model for daily and monthly periods show some substantial differences, whereby policy actions persist in the short-run impact.

In the study of Pesaran et al. (2001), the autoregressive distributive lag (ARDL) model was developed, whereby the dynamics of the independent variables on the dependent variable could be examined in both the short and long horizon. The ARDL model can be implemented whether the variables in a sample are stationary or not, and in long-horizon estimates, we can find appropriate inferences that cannot be attained or inferred when using alternative cointegration analysis techniques (AbuDalu et al., 2014). Quite a few practitioners have used the ARDL model due to several advantages inherent in the model. Nchor and Darkwah (2015) studied the impact of the linkage between exchange rate fluctuations and the nominal interest rate on inflation in Ghana. They claimed that having different orders of integration in the unit roots within three variables can only be investigated using the ARDL model. Ebiringa and Anyaogu (2014) used the ARDL approach to examine the correlation between the exchange rate, inflation, and interest rate. They stated that the ARDL model is a type of integration that allows one to find what result will be attained in the short horizon and if the result will continue in the long horizon.

In most of the reviewed studies, the connection between the interest rate and exchange rate occurs in the short run. The two factors are barely cointegrated over the long run, as is seen in Hnatkovska et al. (2012) and Saraç and Karagöz (2016). On the other hand, prior literature has extensively studied European countries, as well as Organisation for Economic Co-operation and Development (OECD) countries, together with some African countries, such as Nigeria, using traditional methods other than NARDL. Some other recent studies by Alsamara et al. (2017, 2018) also investigated the nonlinearity of the relationship between oil price shocks and money demand and import cost using the same methods of NARDL. The purpose of this study is to try to uncover the missing asymmetric linkage between the interest rate and exchange rate in the sample (ASEAN members), which has not been explored as a whole using the new NARDL method that was recently advanced by Shin et al. (2014).

Model and Methodology

In this study, the ARDL model has been used to justify the linkage between the dependent variable (exchange rate) and the independent variables (interest rate, inflation rate, and broad money supply) using a multivariate model wherein some additional variables have been included. Interest rate is a highly skewed variable, as we can see in Figure A1; thus, in order to decide on the appropriate modeling choice, we transform the interest rate variable into its logarithmic form. Quarterly data vary from one country to another, depending on the availability of data, as seen in Table A1. In order to justify the effects of the other variables, we employ the following multivariate model:

{LnEXR}_{t} = α_{0} + α_{1} {LnIR}_{t} + α_{2} {LnINF}_{t} + α_{3} LnM 2_{t} + ε_{t}

(1)

where EXR denotes the exchange rate (between the ASEAN official domestic currencies and the US dollar), CPI denotes the Consumer Price Index as a measure of the price level, IR denotes the deposit interest rate, and M2 is a measure of the broad money supply. As mentioned in the introductory section, the exchange rate could be positive or negative as a consequence of global developments and globalization, since many businesses nowadays are affected by universal activities. Therefore, the position of the competitiveness of industrial operations, as well as corporations, might be distressed by exchange rate movements.

A few studies have found a negative relationship between the interest rate and exchange rate. Goyal and Arora (2012) used the GARCH model to estimate the effect of interest rate, intervention, and other measures on the level and volatility of exchange rate, using monthly and daily data in India. They found that increasing the interest rate and intervention by the central bank tends to depreciate the Indian currency value in the domestic market and decrease the volatility of the exchange rate. Over the long run, the forecasted rate of a foreign currency’s depreciation is caused by the real interest rate differential, as noted by Hoffmann and MacDonald (2009), who used a VAR bivariate approach to determine the correlation between real interest rate differentials and the real exchange rate. In the short horizon, Chow and Kim (2006) and Hacker et al. (2014) examined the impact of the exchange rate on the interest rate and found that increases in the volatility of the exchange rate are followed by a decline in the variability of the interest rate.

The consistent incline in the overall degree of prices for goods and services is called inflation. Inflation can be estimated by calculating the yearly percentage increase in inflation (CPI). According to Hyder and Shah (2004), higher prices for imported goods are due to a depreciation of the local currency, and eventually, this will dynamically impact the domestic prices through leading to an incline in the price level within a small open economy, which acts as an international price taker. Thus, the expectation of an increase in the overall level of inflation is based upon a fall in the level of the ASEAN domestic currencies when compared to the US dollar. Ebiringa and Anyaogu (2014) studied the relationship among the interest rate, inflation rate, and exchange rate in Nigeria using an ARDL cointegration analysis for the period 1971–2010. Their results showed a positive linkage between the inflation rate and exchange rate. On the other hand, Nucu (2011) found a negative relationship between the exchange rate, money supply, and GDP in Romania.

In this study, both the short-run and long-run cointegration effects are examined. As a result of measuring the short-run impacts, the model in Equation (1), which measures the long-run models and coefficient estimates, only yields long-run effects; consequently, Equation (2) is derived from Equation (1) in order to examine the dynamics in both the short and long run. Therefore, we follow Pesaran et al.’s (2001) bounds testing approach and consider the following error correction model:

\begin{array}{l} Δ {LnEXR}_{t} = α_{0} + \sum_{k = 1}^{n 1} β_{k} Δ {LnEXR}_{t - k} + \sum_{k = 0}^{n 2} δ_{k} Δ {LnIR}_{t - k} + \sum_{k = 0}^{n 3} φ_{k} Δ {LnINF}_{t - k} + \sum_{k = 0}^{n 4} θ_{k} Δ LnM 2_{t - k} \\ + λ_{1} {LnEXR}_{t - 1} + λ_{2} {LnIR}_{t - 1} + λ_{3} {LnINF}_{t - 1} + λ_{4} LnM 2_{t - 1} + μ_{t} \end{array}

(2)

The nonlinear autoregressive distributive lag model estimates the asymmetric impact of the interest rate on the exchange rate and is designed, in particular, for such variables with high volatility, in order to measure the nonlinear effect of interest rate increases and decreases. Two new variables were derived from the interest rate variable in the linear ARDL Equation (2). As mentioned previously, these new variables signify the actions of increasing and decreasing the interest rate unconnectedly. In this process, the sequential analysis technique of the cumulative sum (CUSUM) method is utilized, through which negative values of the interest rate (NEG) and positive values of the interest rate (POS) are introduced as the partial quantity of the negative and positive values, respectively.¹ To illustrate, $NEG = \sum_{j = 1}^{t} Δ {LnIR}_{j}^{-}$ substitutes negative values with 0, and $POS = \sum_{j = 1}^{t} Δ {LnIR}_{j}^{+}$ substitutes positive values with 0. Thus, ${LnIR}_{j}^{-} and {LnIR}_{j}^{+}$ represent the values that are negative and positive, respectively.

\begin{array}{l} Δ {LnEXR}_{t} = α_{0} + \sum_{k = 1}^{n 1} β_{k} Δ {LnEXR}_{t - k} + \sum_{k = 0}^{n 2} δ_{1, k} Δ {POS}_{t - k} + \sum_{k = 0}^{n 3} δ_{2, k} Δ {NEG}_{t - k} \\ + \sum_{k = 0}^{n 4} φ_{k} Δ {LnINF}_{t - k} + \sum_{k = 0}^{n 5} θ_{k} Δ LnM 2_{t - k} + λ_{1} {LnEXR}_{t - 1} + λ_{2} {POS}_{t - 1} + λ_{3} {NEG}_{t - 1} \\ + λ_{4} {LnINF}_{t - 1} + λ_{5} LnM 2_{t - 1} + μ_{t} \end{array}

(3)

Equation (3) represents the NARDL model wherein the POS and NEG variables have replaced the interest rate variable in Equation (2). Shin et al. (2014) demonstrate that Pesaran et al.’s (2001) bounds testing approach could also be applied to Equation (3). Thus, we can determine if there is an existing short-run symmetry or asymmetry between the exchange rate and other independent variables by investigating the measured values from $δ_{1, k}$ and $δ_{2, k}$ . In addition, finding an analogy between $λ_{2}$ and $λ_{3}$ might indicate a long-run asymmetry between the derived variables from the exchange rate (NEG and POS). However, given the relatively small sample size in the present study (48–68 observations), critical values are calculated specific to the sample size (n = 50), in accordance with Narayan (2005).

Empirical Results

One of the ARDL model’s key requirements has to be fulfilled before beginning to analyze the variables within the ARDL and NARDL estimations. The requirement, according to Pesaran et al. (2001), states that the variables can be integrated at level I(0) or at their first difference I(1) but can never be integrated at their second difference I(2). In other words, this can be interpreted as having stationarity in the variables at either level or at their first difference. Thus, the augmented Dickey–Fuller (ADF) test was undertaken for each variable at level with intercept, as well as with intercept and trend, as shown in Table 1. The results in Table 1 confirm no stationary variables in almost all the variables, except for some, such as inflation for Indonesia and the interest rate for Thailand; therefore, there is a unit root in almost all the variables. To solve the problem of unit roots at level, we looked further and estimated the variables in their first difference in Table 1. The results of the ADF test indicate that all variables this time were stationary at the first difference and none of them were integrated at the second level. As a result, we could proceed to report the estimates of models (2) and (3).

Table 1.

The Augmented Dickey–Fuller Test at Level and First-differenced Variables.

			Lexr		Li	Linf	Lm2
Brunei	Intercept		−1.96(1)		−1.117(1)	−2.375(0)	−1.887(0)
	Intercept and trend		−1.58(1)		−1.181(1)	−0.288(0)	−0.6(0)
Cambodia	Intercept		−2.447(10)		−2.481(1)	−1.135(2)	−0.424(1)
	Intercept and trend		−2.442(10)		−2.284(1)	−0.95(2)	−2.232(1)
Indonesia	Intercept		−1.021(1)		−2.746(2)	−2.966(3)*	−1.778(9)
	Intercept and trend		−2.154(1)		−4.499(1)*	−5.65(4)*	1.47(10)
Laos	Intercept		−2.95(10)		−1.51(5)	−2.5795(10)	−0.226(6)
	Intercept and trend		−1.458(10)		−3.0428(3)	−2.96(8)	−1.606(6)
Malaysia	Intercept		−1.083(1)		−2.825(1)***	0.2817(0)	−2.16(2)
	Intercept and trend		−0.734(1)		−2.785(1)	−3.693(1)*	−0.217(5)
Myanmar	Intercept		−0.564(0)		−1.128(1)	−2.019(1)	0.162(1)
	Intercept and trend		−2.017(0)		−1.745(1)	−1.040(3)	−2.74(1)
Philippines	Intercept		−1.463(1)		−1.298(5)	−1.631(1)	0.548(6)
	Intercept and trend		−1.466(1)		−3.504(4)	−0.81(1)	−2.649(2)
Singapore	Intercept		−1.131(2)		−1.985(1)	−0.507(2)	−0.699(1)
	Intercept and trend		−1.1918(2)		−4.015(7)*	−2.1(2)	−1.038(1)
Thailand	Intercept		−1.83(1)		−4.22(10)*	−2.451(5)	−0.695(5)
	Intercept and trend		−0.911(2)		−4.291(10)*	0.401(5)	−2.599(4)
Vietnam	Intercept		−0.87(4)		−1.116(2)	−1.031(2)	−1.80(5)
	Intercept and trend		−2.69(4)		−1.175(2)	−0.885(2)	−3.314(5)***
The augmented Dickey–Fuller test at first difference
Brunei	Intercept	−5.326(0)***		−5.521(0)***		−2.467(4)		−7.154(0)***
	Intercept and trend	−4.973(2)***		−5.519(0)***		−3.525(5)**		−7.494(0)***
Cambodia	Intercept	−3.933(9)***		−5.09(0)***		−6.477(1)***		−5.929(0)***
	Intercept and trend	−4.162(9)***		−5.286(0)***		−6.5347(1)***		−5.881(0)***
Indonesia	Intercept	−5.519(1)***		−4.4032(1)***		−3.898(2)***		−5.728(1)***
	Intercept and trend	−5.558(1)***		−4.408(1)***		−4.774(2)***		−6.922(1)***
Laos	Intercept	−3.1109(2)**		−3.6(5)***		−3.233(2)**		−4.256(5)***
	Intercept and trend	−3.061(9)*		−3.583(5)**		−4.433(2)***		−4.191(5)***
Malaysia	Intercept	−5.461(0)***		−5.313(0)***		−6.084(1)***		−2.410(4)*
	Intercept and trend	−5.732(0)***		−5.277(0)***		−6.048(1)***		−3.241(1)*
		Lexr		Li		Linf		Lm2
Myanmar	Intercept	−7.728(0)***		−5.772(0)***		−4.077(0)***		−4.454(0)***
	Intercept and trend	−7.721(0)***		−5.805(0)***		−4.327(0)***		−4.431(0)***
Philippines	Intercept	−5.532(0)		−3.418(4)***		−4.724(0)***		−4.385(2)***
	Intercept and trend	−5.531(0)***		−3.316(4)*		−4.787(8)***		−4.458(2)***
Singapore	Intercept	−5.821(1)***		−5.281(0)***		−3.331(1)***		−5.378(0)***
	Intercept and trend	−5.83(1)***		−5.449(0)***		−3.305(1)*		−5.335(0)***
Thailand	Intercept	−5.639(1)***		−4.183(1)***		−3.599(4)***		−4.796(2)***
	Intercept and trend	−5.888(1)***		−4.148(1)***		−4.570(4)***		−4.995(2)***
Vietnam	Intercept	−4.368(1)***		−6.853(1)***		−4.508(1)***		−3.649(4)***
	Intercept and trend	−4.340(1)***		−7.002(1)***		−4.577(1)***		−3.813(4)**

Source: The authors.

Notes: The numbers in the parenthesis show the length of the lag utilized in the ADF test (as it has been determined from the set Akaike information criterion [AIC] to a maximum of 6 lags). The critical value at the 1% level for the ADF test with intercept is about −2.87, while the critical value at the 5% level for the ADF test with intercept and trend is about −3.48. *,** and *** indicate significance at 10%, 5%, and 1% levels, respectively.

To estimate the linear ARDL model, we have retained the lagged level of variables and estimated Equation (2) using the Akaike information criterion (AIC) with automatic selection (which automatically selects the number of lags, with maximum order set at 6 for the dependent variable [LnEXR] and 6 for the independent variables [LnIR, LnINF, LnM2]), with constant and trend. The results of the ARDL estimation are shown below in Table 2, where panel A displays the short-run results, panel B displays the long-run results, and panel C displays other diagnostic statistics.

Table 2 shows that in panel A, the interest rate carries a significant short-run effect in only eight countries, namely, Cambodia, Indonesia, Laos, Malaysia, Brunei, Myanmar, Thailand, and Vietnam. However, the Philippines’ and Singapore’s interest rates did not have a significant short-run effect on the exchange rate. On the other hand, the error correction coefficient, named CointEq (1), has to be significant and with a negative sign to support adjustment towards equilibrium (Bahmani-Oskooee & Ardalani, 2006). This holds true for all ASEAN countries except Laos, Malaysia, and Vietnam, which indicates that the disequilibrium of the exchange rate has no speed to adjust back to equilibrium. Moreover, in the case of Brunei, Cambodia, Indonesia, Myanmar, the Philippines, Singapore, and Thailand, the speed of adjustment, whereby the exchange rate returns to equilibrium in the long run, is 68%, 44%, 22%, 68%, 25%, 16%, and 45%, respectively, every quarter (3 months) for the linear ARDL model.

In panel B, five out of the eight countries carry their impact in the long run, and only four out of the five countries have a positive coefficient in the long run (Cambodia, Malaysia, Thailand, Vietnam), and the only country that has a negative coefficient for the interest rate in the long run is Singapore. This implies that an increase in the interest rate will cause an appreciation in the exchange rate of these countries and a depreciation of the exchange rate in Singapore. This phenomenon is due to the fact that higher interest rates tend to reduce inflationary pressures and cause an appreciation in the exchange rate.

On the other hand, the money supply lasted, in the long run, for only four countries, namely, Cambodia, Indonesia, Myanmar (each at the 1% level), and Thailand (at the 10% level). Three out of these four countries have a positive coefficient for the money supply’s effect on the exchange rate in the long run, implying that an increase in the money supply will appreciate the exchange rate (domestic currency per USD), meaning that the domestic currency will appreciate against the US dollar. However, an increase in Myanmar’s money supply will depreciate the exchange rate. That happens because an increase in the money supply leads to an increase in the purchasing power, which affects inflation and creates lower demand on exports.

Lastly, inflation lasted for only two countries in the long run, namely, Indonesia and Myanmar, both at the 1% level. However, inflation for the Philippines and Singapore exhibited a significant impact on the exchange rate in the long run, at the 1% and 5% levels, respectively, although they had no short-run effects. More importantly, only three out of the four countries had a positive coefficient for the impact of inflation on the exchange rate. This indicates that an increase in inflation will cause the exchange rate to appreciate (domestic currency per USD) in the long run for Indonesia, the Philippines, and Singapore. However, an increase in Myanmar’s inflation will cause a depreciation in its exchange rate, due to lower demand on exports.

However, when considering the value of the F-statistics in panel C of Table 2 for the cointegration tests along with the critical values, as suggested by Narayan (2005), the results for Laos, Malaysia, Myanmar, Singapore, Thailand, and Vietnam show that the F-statistic values are greater than the upper bound I(1) at the 10% level, which indicates cointegration between their variables. The results of indicate cointegration in Brunei, Cambodia, Malaysia, Myanmar, Thailand, and Vietnam.

Panel C in Table 2 also reports the results of a number of other diagnostic tests, in order to determine the strength of the applied linear ARDL model. To begin with, the results of the Breusch–Godfrey serial correlation Lagrange multiplier (LM) test indicate the absence of autocorrelation, which is confirmed by the results, since the p-value of the chi-square for all the countries is above the significance level at 5% in all the optimal models, and we may not reject the null hypothesis of no correlation in our model. A separate test, the heteroscedasticity test, indicated that the p-value of the chi-square in all countries is insignificant and above the 5% level; thus, we may not reject the null hypothesis of no heteroscedasticity. In addition, Ramsey’s RESET (regression equation specification error test) statistical test was utilized to test the model for misspecification. As can be seen in panel C, most of the models have insignificant F-statistics, which implies that the majority of the linear ARDL models are properly specified.

In the end, the stability of the long-horizon coefficient is examined along with the short-horizon dynamics. To do so, we follow the path taken by Pesaran and Pesaran (1997) and implement the CUSUM and cumulative sum square (CUSUMQ) tests, as suggested by Brown et al. (1975). All of the coefficient estimates are stable, either individually in the CUSUM or CUSUMQ tests or in both tests, except for Vietnam, which is similar to previous results. Long and Hien (2018) used similar variables to the ones used in this study to estimate the impact of the real effective exchange rate and the deposit rate on the money demand function in Vietnam, and their CUSUM and CUSUMQ tests were also unstable.

Overall, we are unable to find any support for symmetric effects in Brunei, Indonesia, Laos, Myanmar, and the Philippines using the linear ARDL model. However, the lack of support for the symmetric effects could be due to the linear assumption embedded in the linear ARDL model. It is possible that interest rate changes may have asymmetrical effects on the exchange rate and that increases and decreases may have different effects on the exchange rate. To this end, the NARDL model is utilized in the section where the negative interest rate changes and positive interest rate changes are separated, using the partial positive and negative sums introduced in the NARDL model in Equation (3).

Table 2.

Full Information Estimate of the Linear ARDL Model.

Panel A: Short-run results
Variables	Brunei	Cambodia	Indonesia	Laos	Malaysia
∆ $\ln {EXR}_{t - 2}$	1.04 (8.22)	−0.6 (−3.36)	0.33 (−2.77)	−0.44 (−3.25)	−0.23 (−1.81)
∆ $\ln {EXR}_{t - 3}$	−0.50 (−3.64)	0.31 (1.66)		0.20 (2.04)
∆ $\ln {EXR}_{t - 4}$		0.45 (2.41)	−0.29 (−2.53)
∆ $\ln {EXR}_{t - 5}$		−0.44 (−3.75)	0.31 (2.95)
∆ $\ln {IR}_{t - 1}$	−0.05 (−2.74)	0.016 (1.80)	0.53 (4.23)		−0.14 (−4.18)
∆ $\ln {IR}_{t - 2}$		−0.021 (−2.36)	−0.28 (−2.60)	0.03 (3.92)
∆ $\ln {IR}_{t - 3}$				−0.03 (−2.92)
∆ $\ln {IR}_{t - 4}$			−0.25 (−2.40)
∆ $\ln {IR}_{t - 5}$			0.39 (2.93)
∆ $\ln {IR}_{t - 6}$			−0.18 (−2.30)
∆ $\ln M 2_{t - 1}$			1.29 (3.34)
∆ $\ln M 2_{t - 2}$	−0.29 (−3.54)			0.08 (2.14)
∆ $\ln M 2_{t - 3}$				−0.12 (−3.35)
∆ $\ln M 2_{t - 4}$
∆ $\ln M 2_{t - 5}$			−1.31 (−3.56)		0.36 (3.31)
∆ $\ln M 2_{t - 6}$			1.38 (4.78)		−0.45 (−4.11)
∆ $\ln {INF}_{t - 1}$	2.33 (3.67)				−1.40 (−3.14)
∆ $\ln {INF}_{t - 2}$			−0.79 (−2.43)	0.56 (5.01)	3.33 (4.51)
∆ $\ln {INF}_{t - 3}$		−0.044 (−2.65)	−1.32 (3.57)	−0.51 (−3.95)	−1.24 (−2.29)
∆ $\ln {INF}_{t - 4}$		0.04 (2.41)	1.13 (2.98)
∆ $\ln {INF}_{t - 5}$			−1.24 (−3.57)
∆ $\ln {INF}_{t - 6}$				0.12 (1.85)
CointEq (−1)	−0.68 (−5.77)***	−0.44 (−6.13)***	−0.22 (−1.96)*		0.3 (5.55)***
Panel B: Long-run results
ln IR	0.46 (4.65)***	0.453 (6.29)***		−0.26 (−0.45)	2.06 (5.18)***
ln M2	−0.17 (−1.21)	0.012 (0.50)***	−8.09 (−1.94)*	−1.2 (−0.74)	−0.04 (−0.21)
ln INF	0.20 (0.19)	−0.11 (−5.69)	33.14 (1.92)*	5.17 (0.81)	−0.01 (−0.01)
C	0.002 (0.0007)	0.13 (0.21)	−5.64 (−0.47)	9 (0.65)	0.06 (0.1)
Panel C: Other diagnostic test results
F (bounds test)	1.931078	3.647744	2.484464	8.11413	3.965624
$t_{BDM}$	−5.77***	−6.13***	−1.95	−1.58	5.55***
LM	0.2194	0.4847	0.0038	0.8874	0.4015
BP	0.1301	0.1874	0.388	0.4051	0.3196
RESET	0.016	0.1067	0.7717	0.0182	0.945
CUSUM	S	S	S	S	S
CUSUM²	S	U	U	S	U
Panel A: Short-run results
Variables	Myanmar	Philippines	Singapore	Thailand	Vietnam
∆ $\ln {EXR}_{t - 2}$			−0.32 (−2.44)	−0.71 (−3.45)	−0.30 (−2.79)
∆ $\ln {EXR}_{t - 3}$				0.44 (2.02)
∆ $\ln {EXR}_{t - 4}$		−0.20 (−3.10)	0.24 (2.75)	−0.42 (−1.92)	0.50 (3.92)
∆ $\ln {EXR}_{t - 5}$				0.37 (1.81)	−0.50 (−4.75)
∆ $\ln {EXR}_{t - 6}$				−0.23 (−1.74)
∆ $\ln {IR}_{t - 1}$	10.73 (7.39)			−0.04 (−2.95)	0.03 (2.72)
∆ $\ln {IR}_{t - 2}$					−0.07 (−4.63)
∆ $\ln {IR}_{t - 3}$				0.04 (3.03)	0.03 (2.45)
∆ $\ln M 2_{t - 1}$		0.27 (1.92)		−0.96 (−3.47)	−0.04 (−4.98)
∆ $\ln M 2_{t - 2}$
∆ $\ln M 2_{t - 3}$				−0.37 (−2.40)
∆ $\ln M 2_{t - 4}$					0.04 (4.43)
∆ $\ln M 2_{t - 5}$					−0.04 (−3.34)
∆ $\ln M 2_{t - 6}$					0.01 (1.71)
∆ $\ln {INF}_{t - 1}$		−1.16 (−2.54)
∆ $\ln {INF}_{t - 2}$	3.46 (3.68)	1.47 (3.20)	−0.87 (−3.08)
∆ $\ln {INF}_{t - 3}$					0.31 (3.74)
∆ $\ln {INF}_{t - 4}$					−0.46 (−3.52)
∆ $\ln {INF}_{t - 5}$					0.37 (3.10)
∆ $\ln {INF}_{t - 6}$					−0.29 (−4.04)
CointEq (−1)	−0.68 (−9.6)***	−0.25 (−3.19)***	−0.16 (−2.19)**	−0.45 (−4.69)***	0.3 (5.85)***
Panel B: Long-run results
ln IR	0.33 (0.25)	0.17 (1.61)	−2.09 (−2.21)**	0.13 (3.46)***	0.95 (5.83)***
ln M2	21.63 (9)***	−0.23 (−1.41)	0.02 (0.09)	0.34 (1.93)*	0.03 (1.12)
ln INF	−13.82 (−8.39)***	2.55 (5.04)***	4.88 (2.37)**	−0.21 (−0.37)	0.04 (0.7)
C	−0.75 (−0.32)	−7.75 (−3.65)***	−4.89 (−3.2)***	−0.73 (−1.41)	−0.47 (−0.8)
Panel C: Other diagnostic test results
F (bounds test)	30.72399	3.12324	4.159875	4.035934	7.688157
LM	0.8211	0.6337	0.0412	0.4523	0.5435
$t_{BDM}$	−9.60***	−3.19	−2.19	−4.68***	11.18***
BP	0.5766	0.6655	0.1805	0.4489	0.5258
RESET	0.0013	0.3044	0.2095	0.9807	0.247
CUSUM	S	S	S	S	U
CUSUM²	U	S	S	S	U

Source: The authors.

Notes: The model selection criteria for the ARDL model uses Akaike information criterion (AIC) to a maximum of 6 lags with automatic lag selection. Here, we follow the general-to-specific approach to select the final ARDL specification with a maximum of 6 lags, dropping all insignificant stationary regressors, following the method of Shin et al. (2014). The critical value of the F-test for cointegration when there are three exogenous variables (k = 3, n = 50) is 3.973 (4.700) for the upper bound and 2.873 (3.500) for the lower bound at the 10% (5%) level of significance (found in Narayan, 2005, Appendix, Case III, p. 1988). The critical value of $t_{bdm}$ for cointegration when there are three exogenous variables is 3.45. Numbers inside the parentheses next to coefficient estimates are absolute value of t-ratios. *,** and *** indicate significance at 10%, 5%, and 1% levels, respectively.

The significance of CUSUM tests is attached in the appendix in Figure A2.

By estimating the nonlinear ARDL, we retain the lagged level of variables and estimate Equation (3) as shown in Table 3. In panel A, since at least either one of the ${POS}^{IR}$ and ${NEG}^{IR}$ coefficients is significant in all countries, it can be inferred that the interest rate has an asymmetric effect in the short run in all countries. For the other variables (M2 and INF), the money supply seems to be significant in all countries other than Singapore and Brunei. However, inflation is only significant in six countries (Brunei, Cambodia, Laos, Malaysia, the Philippines, and Singapore). Moreover, the coefficients of M2 and INF have the expected signs in most countries in the NARDL in the short run. According to Shin et al. (2014), it is recommended to test for short-run “impact asymmetry” by implementing the Wald test to investigate whether the sum of the short-run coefficients attached to ΔPOS^IR is different from the sum of the short-run coefficients attached to ΔNEG^IR. This is shown in panel C in Table 2, denoted by Wald-S. The short-run results seem to show evidence of a significant short-run cumulative or impact asymmetry in all the countries. Do these short-run effects continue in the long run?

The results of the tests to check whether the short-run effects of the ${POS}^{IR}$ and ${NEG}^{IR}$ coefficients last in the long run are reported in Table 3, panel B. We can infer that only seven countries hold their asymmetric effects in the long run (Brunei, Cambodia, Indonesia, Myanmar, the Philippines, Singapore, and Thailand). However, the asymmetric effects of the interest rate in Malaysia and Vietnam do not transform into the long run. Consequently, it seems that presenting a nonlinear adjustment of the interest rate increases the number of countries in which either the increase or decrease, or both, can have a significant long-run impact on the interest rate. How do we know that these significant long-run effects vary from one another? To show the variance among these long-run estimations and to deliver a statistical indication for the long-run asymmetric effects, we implement the Wald test. This Wald statistic is denoted as Wald-L in panel C, and it is significant in the results shown for Indonesia, Myanmar, the Philippines, and Singapore.

As for the other diagnostic statistical tests in panel C, considering the value of the F-statistic for the cointegration tests, along with the critical values as suggested by Narayan (2005), the results for all countries except Malaysia show that the F-statistic values are greater than the upper bound I(1) at the 10% level, which indicates cointegration between their variables. However, the results of

t_{BDM}

indicate cointegration in Cambodia, Singapore, and Thailand. Moreover, the absence of autocorrelation is confirmed, as the p-value of the chi-square for all the countries is above the significance level at 5% in all optimal models; thus, we may not reject the null hypothesis of no correlation in our model. The presence of heteroskedasticity is tested using the Breusch–Pagan–Godfrey test. Since the p-value of the chi-square in all the countries is insignificant and above the 5% level, we may not reject the null hypothesis of no heteroscedasticity, except in the case of Myanmar. The Ramsey RESET statistics have insignificant F-statistics in most of the models; this implies that the majority of the linear ARDL models are properly specified, except in the case of Myanmar. Finally, all estimated coefficients seem to be stable, as reflected in the results of the CUSUM and CUSUMSQ tests, except for the CUSUMSQ results for Indonesia and Malaysia.

Table 3.

Full Information Estimate of the Nonlinear ARDL (NARDL) Model.

Panel A: Short-run results
Variables	Brunei	Cambodia	Indonesia	Laos	Malaysia
∆ $\ln {EXR}_{t - 2}$	−0.34 (−2.78)	−0.47 (−3.53)	−0.38 (−3.33)	−0.59 (−3.92)
∆ $\ln {EXR}_{t - 3}$		0.33 (2.41)	0.33 (3.07)	0.36 (3.49)	−0.38 (−3.30)
∆ $\ln {EXR}_{t - 4}$		0.63 (4.95)	−0.27 (−2.37)
∆ $\ln {EXR}_{t - 5}$		−0.70 (−7.33)	0.38 (4.93)		0.41 (2.70)
∆ $\ln {EXR}_{t - 6}$					−0.35 (−2.59)
∆ $l n {POS}_{t - 2}^{IR}$	−0.11 (−2.33)	−0.08 (−5.43)	0.34 (8.69)		−0.14 (−1.89)
∆ ${lnPOS}_{t - 3}^{IR}$	0.13 (2.55)				0.30 (3.03)
∆ ${lnPOS}_{t - 4}^{IR}$		0.08 (6.18)			−0.19 (−3.22)
∆ ${lnPOS}_{t - 5}^{IR}$	0.12 (2.13)
∆ ${lnPOS}_{t - 6}^{IR}$	−0.26 (−4.75)
∆ ${lnNEG}_{t - 1}^{IR}$					−0.16 (−3.08)
∆ ${lnNEG}_{t - 2}^{IR}$		−0.03 (−3.44)		0.03 (2.79)	−0.17 (−3.98)
∆ ${lnNEG}_{t - 3}^{IR}$	0.078 (3.15)	0.05 (3.54)		−0.06 (−3.65)
∆ ${lnNEG}_{t - 4}^{IR}$		−0.04 (−2.77)	−0.37 (−4.25)		0.11 (2.52)
∆ ${lnNEG}_{t - 5}^{IR}$		−0.02 (−1.92)	0.51 (3.77)
∆ ${lnNEG}_{t - 6}^{IR}$		0.03 (3.55)	−0.51 (−5.72)
∆ $\ln M 2_{t - 1}$		0.03 (3.12)
∆ $\ln M 2_{t - 2}$		−0.05 (−5.12)	−0.45 (−2.09)		−0.24 (−2.01)
∆ $\ln M 2_{t - 3}$				−0.05 (−4.00)	0.48 (4.12)
∆ $\ln M 2_{t - 4}$			0.46 (2.28)
∆ $\ln M 2_{t - 5}$		0.05 (5.19)
∆ $\ln M 2_{t - 6}$		−0.03 (−3.95)	0.67 (3.88)		−0.29 (−4.12)
∆ $\ln {INF}_{t - 1}$					−2.21 (−4.76)
∆ $\ln {INF}_{t - 2}$				0.43 (3.49)	4.41 (6.17)
∆ $\ln {INF}_{t - 3}$	−3.12 (−4.53)	−0.02 (−2.82)		−0.19 (−1.66)
∆ $\ln {INF}_{t - 4}$					−1.72 (−2.65)
∆ $\ln {INF}_{t - 5}$					−1.42 (−2.10)
∆ $\ln {INF}_{t - 6}$	2.27 (3.86)				1.46 (2.45)
CointEq (−1)	−0.36 (−1.08)	−0.27 (−3.58)***	−0.22 (−2.29)**	−0.04 (−0.91)	−0.11 (−0.92)
Panel B: Long-run results
${lnPOS}^{IR}$	−0.22 (−1.93)*	0.06 (0.67)	−0.92 (−0.92)	−0.31 (−0.43)	−0.76 (−0.41)
${lnNEG}^{IR}$	0.11 (0.32)	−0.05 (−2.00)*	−3.23 (−3.23)***	0.73 (0.94)	−0.39 (−0.34)
ln M2	0.66 (0.33)	−0.01 (−0.54)	−1.95 (−1.95)*	−5.33 (−0.88)	−0.69 (−0.37)
ln INF	2.21 (0.41)	−0.04 (−1.29)	−1.25 (−1.25)	9.81 (0.89)	7.81 (0.43)
C	−16.54 (−1.38)	−3.34 (−17.58)***	2.12 (−2.12)**	−10.19 (−1.5)	−6.54 (−0.59)
Panel C: Other diagnostic test results
F (bounds test)	5.102438	6.303828	14.60234	3.750006	2.21044
$t_{BDM}$	−1.079	−3.57*	−2.29	−1.53	−0.92
LM	0.1258	0.1869	0.4739	0.1856	0.0644
BP	0.7225	0.5309	0.2342	0.4941	0.9006
RESET	0.558	0.0548	0.1518	0.0571	0.2823
CUSUM	S	S	S	S	S
CUSUM²	S	S	U	S	U
Wald-S	7.58***	18.8***	20.91***	37.16***	13.5***
Wald-L	1.38	0.2	3.16*	2.27	0.91
Panel A: Short-run results
Variables	Myanmar	Philippines	Singapore	Thailand	Vietnam
∆ $\ln {EXR}_{t - 2}$	−0.69 (−3.94)			−0.78 (−4.14)
∆ $\ln {EXR}_{t - 3}$	0.99 (4.05)		−0.28 (−2.48)	0.48 (2.44)
∆ $\ln {EXR}_{t - 4}$				−0.39 (−1.96)	0.84 (7.51)
∆ $\ln {EXR}_{t - 5}$				0.44 (2.43)	−0.43 (−3.97)
∆ $\ln {EXR}_{t - 6}$				−0.28 (−2.32)	−0.24 (−2.91)
∆ ${lnPOS}_{t - 1}^{IR}$			0.38 (3.52)		−0.05 (−5.31)
∆ ${lnPOS}_{t - 2}^{IR}$			−0.32 (−4.35)	−0.08 (−2.16)
∆ ${lnPOS}_{t - 3}^{IR}$				0.13 (3.62)	−0.07 (−3.08)
∆ln ${POS}_{t - 4}^{IR}$					0.09 (4.14)
∆ ${lnNEG}_{t - 1}^{IR}$	22.61 (9.33)			−0.07 (−2.78)	0.06 (4.74)
∆ ${lnNEG}_{t - 2}^{IR}$	−15.01 (−3.54)				−0.06 (−5.01)
∆ ${lnNEG}_{t - 3}^{IR}$	27.32 (4.40)	−0.09 (−3.44)
∆ ${lnNEG}_{t - 4}^{IR}$	−22.14 (−4.03)	0.06 (2.42)			−0.02 (−1.75)
∆ ${lnNEG}_{t - 5}^{IR}$
∆ $\ln M 2_{t - 1}$		0.30 (2.21)		−0.91 (−3.96)	−0.05 (−6.07)
∆ $\ln M 2_{t - 2}$	−0.26 (−2.96)
∆ $\ln M 2_{t - 3}$				−0.39 (−2.62)	−0.02 (−2.58)
∆ $\ln M 2_{t - 4}$					0.02 (2.81)
∆ $\ln M 2_{t - 5}$					−0.03 (−4.63)
∆ $\ln {INF}_{t - 1}$		−1.03 (−2.18)
∆ $\ln {INF}_{t - 2}$		1.15 (2.05)	−0.79 (−3.19)
∆ $\ln {INF}_{t - 4}$		0.65 (2.27)
CointEq (−1)	−0.46 (−2.48)**	−0.2 (−3.11)***	−0.36 (−3.81)***	−0.45 (−4.9)***	0.301 (1.932)*
Panel B: Long-run results
${lnPOS}^{IR}$	0.19 (0.33)	−0.45 (−2.85)***	−0.36 (−4.13)***	0.13 (2.45)**	0.25 (1.55)
${lnNEG}^{IR}$	11.53 (25.24)***	0.14 (1.22)	0.01 (0.26)	−0.04 (−2.01)*	−0.02 (−0.24)
ln M2	−0.58 (−3.18)***	−0.36 (−1.58)	0.67 (4.56)***	−0.55 (−5.83)***	0.11 (3.96)***
ln INF	0.73 (2.14)**	4.2 (4.85)***	−0.67 (−1.92)**	0.64 (1.1)	−0.34 (−1.91)*
C	1.69 (2.28)**	−7.01 (−3.51)***	−2.45 (−4.7)***	0.13 (2.45)	0.02 (0.01)
Panel C: Other diagnostic test results
F (bounds test)	3.964578	5.728199	4.602423	6.757986	5.884907
$t_{BDM}$	−2.47	−3.10	−3.80*	−4.89***	1.93
LM	0.2928	0.0519	0.3462	0.7263	0.3153
BP	0.0027*	0.7735	0.2157	0.291	0.7883
RESET	0.00	0.5352	0.1952	0.3747	0.0447
CUSUM	S	S	S	S	S
CUSUM²	S	S	S	S	S
Wald-S	46.07***	75.25***	25.51***	25.00***	15.49***
Wald-L	4.76**	10.61***	5.10**	0.9	0.11

Source: The authors.

Numbers inside the parentheses next to coefficient estimates are absolute value of t-ratios. *,** and *** indicate significance at 10%, 5%, and 1% levels, respectively.

To sum up, the results of the linear ARDL model depict that the interest rate’s effects on the exchange rate is carried into the long run in only five ASEAN member countries, namely, Cambodia, Malaysia, Singapore, Thailand, and Vietnam. However, in the NARDL model, the interest rates in seven countries, namely, Brunei, Cambodia, Indonesia, Myanmar, the Philippines, Singapore, and Thailand, have nonlinear asymmetric effects on the respective exchange rates, while Laos reported no long-run relationship in either models. On the other hand, the results of the short-run ARDL and NARDL models vary a little. In the short run, in the ARDL model, only Brunei and Myanmar have no short-run effects, while in the NARDL model, only Laos has no nonlinear asymmetric effects. Interestingly, Laos only has a short-run effect of its interest rate in the linear ARDL model.

Comparing the linear ARDL and NARDL models in the short run, we conclude that the interest rates of seven countries have both symmetric and asymmetric effects on the respective exchange rates, namely, Cambodia, Indonesia, Malaysia, the Philippines, Singapore, Thailand, and Vietnam. Laos only has a symmetric effect, and Myanmar and Brunei only have asymmetric effects, in the short run. This implies that all ASEAN countries have asymmetric effects in the short run, except for Laos. The negative changes in the interest rate in Myanmar and positive changes in the interest rate in Singapore were found to affect the exchange rate in the short run. However, both positive and negative changes in the interest rates in Brunei, Cambodia, Indonesia, Malaysia, the Philippines, Thailand, and Vietnam were found to have short-run effects on the exchange rate. Comparing the short-run linear ARDL model and the long-run ARDL model, it can be seen that the symmetric short-run effects do not last in the long run in all ASEAN members. However, in the case of Cambodia, Malaysia, Singapore, Thailand, and Vietnam, these effects last in the long run. The results of the linear ARDL model show that the statistical coefficient related to LnIR is greater in the long run than in the short run, indicating a more significant symmetric effect of the interest rate on the exchange rate in the long run compared to the short run.

On the other hand, in the NARDL model, we can infer that the asymmetric short-run effect only transforms into the long run in Brunei, Cambodia, Indonesia, Myanmar, the Philippines, Singapore, and Thailand. By benchmarking the absolute values related to the coefficients of the ${POS}^{IR}$ and ${NEG}^{IR}$ variables, we can infer a larger asymmetric effect of interest rate changes in the short run, compared to the long run, in Brunei, Myanmar, and Thailand, whereas a larger asymmetric effect of interest rate changes in the long run, compared to the short run, is inferred in Indonesia, the Philippines, and Singapore. However, in Cambodia, the asymmetric effects seem to have the same effect in both the short and long run.

In addition, evidence of significant long-run asymmetric effects has been found in Indonesia, Myanmar, and the Philippines, whereby positive and negative changes in the interest rate are not only supported by the NARDL model but also by the Wald test. If we had only considered the linear ARDL model, we would not have uncovered any relationship in the long run for these countries. For instance, in the case of Indonesia, if we had only looked into the linear ARDL model, we would have concluded that there were no long-run effects and stopped there, since the coefficient of the interest rate in the linear ARDL model in the long run is insignificant. After all, increases and decreases in the interest rate have significant long-run asymmetric effects in Indonesia, Myanmar, and the Philippines, which were concealed by the linear model. Therefore, taking the nonlinear modification of the interest rates in Indonesia, Myanmar, and the Philippines yields a different result that was hidden by the linear ARDL model. For instance, the long-run asymmetric impacts in Myanmar are not the same as those in Indonesia. While in Indonesia a decrease in the interest rate leads to currency depreciation, it is the other way round in Myanmar.

Conclusion

Studies on the linkage between the interest rate and exchange rate indicate a relationship between them. One of the economic models, known as the Mundell–Fleming model, proposes that in the short run, a decrease in the domestic country’s interest rates causes a deficiency in the balance of payments which can only be handled by depreciating the home country’s currency through an increase in net exports. On the other hand, when an increase in the domestic interest rate occurs, it attracts more foreign investment and leads to an appreciation of the domestic currency, due to a rise in the exchange rate. Therefore, the exchange rate can move in either direction unless the interest rate is monitored in order to thrive in a certain economy. One case of concern in the East Asian region is the exchange rates of the ASEAN countries.

The connection between the interest rate and exchange rate that is established in the literature is rich and incorporates numerous investigations. These examinations have two basic highlights. The first is the conclusion that the connection between an interest rate and the corresponding exchange rate occurs in the short run. The two factors are barely cointegrated over the long run. Second, the focus of the literature extensively is on European and OECD countries, and some African countries, such as Nigeria, using traditional methods, rather than the NARDL technique.

The relationship between the interest rate and exchange rate is crucial; however, the deficiency of clear empirical evidence of the effect of the interest rate on the exchange rate is without exception more ambivalent from the perception of an economist. Thus, in this study, we aimed to uncover whether interest rate changes have symmetric or asymmetric effects on exchange rate volatility in ASEAN countries and compare this linkage among them. The significance of meeting the wished-for research objectives and determining the camouflaged interest rate–exchange rate linkage is due to the possible actions of policymakers in using interest rate mechanisms to manipulate the exchange rate. Thus, we considered the linear ARDL model to study the impacts of the interest rate on the exchange rate; however, the increases and decreases of the interest rates have varying effects on the exchange rate, and that is why we extended the study by adopting the nonlinear model. Keeping that in mind, the linear ARDL and bounds testing methodology of Pesaran et al. (2001) was utilized to affirm the past literature. Following that, the NARDL approach was utilized, which was introduced in recent times by Shin et al. (2014) to uncover the symmetric or asymmetric impacts of interest rate changes on the exchange rate. Furthermore, the article followed Pesaran et al.’s (2001) bounds testing approach and used the error correction model.

Quarterly data of each of the 10 ASEAN members were used to investigate the symmetric and asymmetric cointegration effects, which are listed in Table A1. Before estimating the ARDL and NARDL models, the data were checked for existence of unit root. Unit root was found at level but not at the first difference, whereby the null hypothesis of having nonstationary variables was rejected. Subsequently, we applied the linear and nonlinear ARDL models, and conclusions were made with regard to the results. The results of the long-run ARDL model showed that the interest rate of only six ASEAN members impacted the exchange rate. However, in the nonlinear ARDL model, the interest rate of seven countries had nonlinear asymmetric effects on their exchange rate, while Laos reported no long-run relationship in either model. Consequently, it seems that there was an indication of long-run asymmetric effects of interest rate changes on the exchange rate in more countries in the NARDL model compared to the linear ARDL model. The other considered variables (money supply and inflation) showed either a similar behavior, in some countries, or a different behavior, in other countries, in the long run. A similar behavior can be inferred in the money supply of Indonesia, Myanmar, and Thailand, as well as in the inflation in Myanmar, the Philippines, and Singapore, as they appeared to have significant effects in both the linear ARDL model and the NARDL model, indicating that these countries should pay more attention when their policymakers implement new policies to manipulate their exchange rate. On the other hand, a different behavior was observed in the money supply of Cambodia, as it only had a significant effect in the long run according to the ARDL model. Additionally, in the long run, inflation in Indonesia had a significant effect in the linear ARDL model, and inflation in Vietnam had a significant effect according to the NARDL model.

A crucial policy implication of our results is that ASEAN members that try to increase their interest rates with the prospect of increasing foreign investment should be fully aware of the asymmetric effects of their actions on their exchange rate. Policymakers in ASEAN countries should take into consideration the effects of the interest rate on the exchange rate, as the former turns out to have various effects on the exchange rate. Our findings show that, with regard to short-run asymmetric effects, increases in the interest rate impacted the exchange rate positively in Brunei, whereas they had negative effects on the rest of the ASEAN members, with the exception of Laos.

When extending the linear ARDL to the NARDL model, our study on the impact of the interest rate on the exchange rate revealed the masked long-run asymmetric cointegration, which must be taken into consideration in future research. Furthermore, studying the impact of exchange rate changes on the interest rate could be a possibility for new research. This study only looked into the impact of the interest rate on the exchange rate. In addition, other variables, such as inflation rate and industrial production index, could be included in the NARDL model to fix the misspecification of the studied model, as noted in the case of Myanmar.

Appendix

Data Source

The impact of macroeconomic variables on the exchange rate among ASEAN countries has been examined using secondary data. The data were collected from the International Monetary Fund (www.imf.org/en/Data) for all the variables; however, certain variables of some countries, such as Thailand’s deposit rate and Vietnam’s money supply, were collected from the International Financial Statistics, found at http://stats.ukdataservice.ac.uk. Moreover, the money supply and deposit rates of Laos were collected from the Bank of Lao PDR (http://www.bol.gov.la). The period of the data varies from one country to another depending on the availability, as listed in Table A1.

Table A1.

Collected Data Timeframe.

Country	Data Period	Observation Number
Brunei Darussalam	2004 Q1–2017 Q3	53
Cambodia	2002 Q1–2017 Q3	59
Indonesia	2004 Q3–2017 Q3	48
Laos	2000 Q1–2016 Q4	64
Malaysia	2002 Q1–2017 Q2	58
Myanmar	2002 Q1–2017 Q3	59
The Philippines	2002 Q1–2017 Q2	62
Singapore	2000 Q1–2017 Q4	68
Thailand	2002 Q1–2017 Q2	62
Vietnam	2002 Q1–2016 Q4	56

Source: International Monetary Fund, International Financial Statistics, Bank of Lao PDR.

Figure A1.

Source: World Development Indicators.

Figure A2.

Source: The authors.

Footnotes

Declaration of Conflicting Interests

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research received no specific funding from any funding agency in the public, commercial, or not-for-profit sectors.

Note

References

Abbas

Iqbal

Ayaz

(2012). Relationship between GDP, inflation and real interest rate with exchange rate fluctuation of African countries. International Journal of Academic Research in Accounting, Finance and Management Sciences , 2(3), 132–141.

AbuDalu

Ahmed

E. M.

Almasaied

S. W.

Elgazoli

A. I.

(2014). The real effective exchange rate impact on Asean-5 economic growth. International Journal of Economics and Management Sciences , 3(174), 2.

Agénor

P. R.

McDermott

C. J.

Üçer

(1997). Fiscal imbalances, capital inflows, and the real exchange rate: The case of Turkey. European Economic Review , 41(3–5), 819–825.

Akçağlayan

(2008). The effects of interest rate policy on exchange rates during 2001 currency crises. Abant Izzet Baysal University Journal of Social Sciences , 9(1), 1–20.

Alsamara

Mrabet

Dombrecht

(2018). Asymmetric import cost pass-through in GCC countries: Evidence from nonlinear panel analysis. Economic Modelling , 75, 432–440.

Alsamara

Mrabet

Dombrecht

Barkat

(2017). Asymmetric responses of money demand to oil price shocks in Saudi Arabia: A non-linear ARDL approach. Applied Economics , 49(37), 3758–3769.

Apergis

Miller

(2006). Consumption asymmetry and the stock market: Empirical evidence. Economics Letters , 93(3), 337–342.

Asari

F. F. A. H.

Baharuddin

N. S.

Jusoh

Mohamad

Shamsudin

Jusoff

(2011). A vector error correction model (VECM) approach in explaining the relationship between interest rate and inflation towards exchange rate volatility in Malaysia. World Applied Sciences Journal , 12(3), 49–56.

Bahmani-Oskooee

Ardalani

(2006). Exchange rate sensitivity of US trade flows: Evidence from industry data. Southern Economic Journal , 72(3), 542–559.

10.

Bahmani-Oskooee

Bahmani

(2015). Nonlinear ARDL approach and the demand for money in Iran. Economics Bulletin , 35(1), 381–391.

11.

Bahmani-Oskooee

Fariditavana

(2014). Do exchange rate changes have symmetric effect on the S-curve? Economics Bulletin , 34(1), 164–173.

12.

Berument

(2007). Measuring monetary policy for a small open economy: Turkey. Journal of Macroeconomics , 29, 411–430.

13.

Bjørnland

H. C.

(2009). Monetary policy and exchange rate overshooting: Dornbusch was right after all. Journal of International Economics , 79(1), 64–77.

14.

Branson

W. H.

(1981). Macroeconomic determinants of real exchange rates (NBER Working Papers 0801). https://www.nber.org/papers/w0801

15.

Branson

W. H.

Halttunen

(1979). Asset-market determination of exchange rates: Initial empirical and policy results. In Martin

J. P.

Smith

(Eds.), Trade and payments adjustment under flexible exchange rates (pp. 55–85). Macmillan.

16.

Branson

W. H.

Halttunen

Masson

(1977). Exchange rates in the short run: The dollar-dentschemark rate. European Economic Review , 10(3), 303–324.

17.

Brown

R. L.

Durbin

Evans

J. M.

(1975). Techniques for testing the constancy of regression relationships over time. Journal of the Royal Statistical Society: Series B (Methodological) , 37(2), 149–163.

18.

Chow

H. K.

Kim

(2006). Does greater exchange rate flexibility affect interest rates in post-crisis Asia? Journal of Asian Economics , 17(3), 478–493.

19.

Clarida

Gali

(1994, December). Sources of real exchange-rate fluctuations: How important are nominal shocks? Carnegie-Rochester Conference Series on Public Policy , 41, 1–56.

20.

Delatte

A.-L.

Lopez-Villavicencio

(2012). Asymmetry exchange rate pass-through: Evidence from major countries. Journal of Macroeconomics , 34(3), 833–844.

21.

Ebiringa

O. T.

Anyaogu

N. B.

(2014). Exchange rate, inflation and interest rates relationships: Autogressive distributive lag analysis. Journal of Economics and Development Studies , 2, 263–279.

22.

Engle

R. F.

Bollerslev

(1986). Modelling the persistence of conditional variances. Econometric Reviews , 5(1), 1–50.

23.

Goyal

Arora

(2012). The Indian exchange rate and Central Bank action: An EGARCH analysis. Journal of Asian Economics , 23(1), 60–72.

24.

Gül

Ekinci

Özer

(2007). Causal relationship between interest rates and exchange rate in Turkey: 1984–2006. İktisat, İşletmeveFinans , 251, 21–31.

25.

Gümüş

İ.

(2002). Research and monetary policy department, effects of the interest rate defense on exchange rates during the 1994 crisis in Turkey (Working Papers 0214). https://www.tcmb.gov.tr/wps/wcm/connect/a011f489-8522-4f35-984b-aa046c524e7b/17.pdf?MOD=AJPERES&CACHEID=ROOTWORKSPACE-a011f489-8522-4f35-984b-aa046c524e7b-m3fw5Hr

26.

Hacker

R. S.

Karlsson

H. K.

Månsson

(2014). An investigation of the causal relations between exchange rates and interest rate differentials using wavelets. International Review of Economics & Finance , 29, 321–329.

27.

Hnatkovska

V. V.

Lahiri

Vegh

C. A.

(2012). The exchange rate response puzzle . https://dx-doi-org-s.web.bisu.edu.cn/10.2139/ssrn.1996693

28.

Hnatkovska

Lahiri

Vegh

(2013). Interest rate and the exchange rate: A non-monotonic tale. European Economic Review , 63(C), 68–93.

29.

Hoffmann

MacDonald

(2009). Real exchange rates and real interest rate differentials: A present value interpretation. European Economic Review , 53(8), 952–970.

30.

Holland

Domowitz I.

Hakkio

C. S.

(1985). Conditional variance and the risk premium in foreign exchange market. Journal of International Economics , 19(1–2), 47–66.

31.

Huang

Hueng

J. M.

&Yau

(2010). Traditional view or revisionist view? The effects of monetary policy on exchange rates in Asia. Applied Financial Economics , 20(9), 753–760.

32.

Hyder

Shah

(2004). Exchange rate pass- through to domestic prices in Pakistan (SBP Working Paper Series 05). State Bank of Pakistan, Research Department.

33.

Kayhan

Bayat

Ahmet

U. Ğ. U. R.

(2013). Interest rates and exchange rate relationship in BRIC-T countries. Ege Akademik Bakış Dergisi , 13(2), 227–236.

34.

Kisaka

S. E.

Kithitu

J. W.

Kamuti

H. M.

(2014). The causal relationship between interest rates and foreign exchange rates in Kenya. Research Journal of Finance and Accounting , 5(14), 136–151.

35.

Lily

Kogid

Karim

M. R. A.

Asid

Mulok

(2011). Empirical testing on uncovered interest rate parity in Malaysia. Journal of Applied Finance and Banking , 1(2), 95.

36.

Long

P. D.

Hien

B. Q.

(2018 January). Determinants and stability of demand for money in Vietnam . In International Econometric Conference of Vietnam, pp. 712–726. Springer, Cham.

37.

MacDonald

Nagayasu

(2000). The long-run relationship between real exchange rates and real interest rate differentials: A panel study. IMF Staff Papers , 47(1), 116–128.

38.

Narayan

(2005). The saving and investment nexus for China: Evidence from cointegration tests. Applied Economics , 37(17), 1979–1990.

39.

Nchor

Darkwah

S. A.

(2015). Inflation, exchange rates and interest rates in Ghana: An autoregressive distributed lag model. Acta Universitatis Agriculturae et Silviculturae Mendelianae Brunensis , 63(3), 969–977.

40.

Nucu

A. E.

(2011). The relationship between exchange rate and key macroeconomic indicators. Case study: Romania. The Romanian Economic Journal , 41, 127–145.

41.

Perasan

M. H.

Pesaran

(1997). Working with Microfit 4.0: Interactive econometric analysis . Oxford University Press.

42.

Pesaran

M. H.

Shin

Smith

R. J.

(2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied Econometrics , 16(3), 289–326.

43.

Rasband

S. N.

(1990). Chaotic dynamics of non-linear systems . Wiley, New York.

44.

Saraç

T. B.

Karagöz

(2016). Impact of short-term interest rate on exchange rate: The case of Turkey. Procedia Economics and Finance , 38, 195–202.

45.

Shin

Greenwood-Nimmo

(2014). Modelling asymmetric cointegration and dynamic multipliers in a nonlinear ARDL framework. In Sickles

Horrace

W. C.

(Eds.), Festschrift in honor of Peter Schmidt (pp. 281–314). Springer.

46.

Sin

(2016 May 4). ASEAN to become world’s 4th largest economy by 2030. The Jakarta Post . http://www.thejakartapost.com/seasia/2016/05/04/asean-to-become-worlds-4th-largest-economy-by-2030.html

47.

Sollis

Wohar

M. E.

(2006). The real exchange rate–real interest rate relation: Evidence from tests for symmetric and asymmetric threshold cointegration. International Journal of Finance & Economics , 11(2), 139–153.

48.

Tari

Abasiz

(2009). Frequency domain approach and short run and long run causality test: Evidence from Turkey for interest rate and exchange rate relationship. METU Studies in Development , 36, 405–421.

49.

The Star. (2017). Asean needs mechanism to address forex volatility - economist - Business News | The Star Online. (2017). Retrieved from: http://www.thestar.com.my/business/business-news/2017/09/15/asean-needs-mechanism-to-address-forex-volatility—economist/

50.

Verheyen

(2013). Interest rate pass-through in the EMU: New evidence using nonlinear ARDL framework. Economics Bulletin , 33(1), 729–739.

51.

World development indicators. The World Bank: Washington, D.C.