Exchange rates can be defined as the value of a domestic currency in terms of one or more international currencies ( Moosa & Bhatti, 2010 ). Bilateral exchange rates, representing the exchange between two currencies, can be measured either using a nominal measure or a real (or price-level adjusted) measure ( Evans, 2011 ). Exchange rates can also be expressed as an effective exchange rate, which represents the exchange rate of a domestic currency against a basket of exchange rates, typically including major trading partners ( MacDonald, 2007 ). Nominal effective exchange rates (NEER) are calculated without price level adjustment, while real effective exchange rates (REER) are calculated using price level adjustment ( MacDonald, 2007 ).

Exchange rate volatility is defined as the interperiod variance in exchange rates within a currency pair ( Gali & Monacelli, 2005 ). Exchange rate volatility is typically associated with exchange rate regimes; fixed exchange rate policies are associated with low levels of exchange rate volatility, while intermediate and floating regimes are associated with higher exchange rate volatility levels ( Frieden, 2015 ). Inflation-targeting regimes, which adjust exchange rates in order to control domestic inflation, are also associated with higher exchange rate volatility ( Ouyang et al., 2016 ). Exchange rate volatility also varies across time horizons, even in the absence of policy changes, with short-run volatility typically being more predictable than long-run volatility ( Marston, 1987 ). There are also local conditions that can play a role in exchange rate volatility, particularly dual exchange rate policies, heavy trading in informal exchange, and dollarized economies ( Evans, 2011 ). Thus, while there are some generally well-understood causes of exchange rate volatility, there can also be a lot of variation between countries and time periods. Furthermore, the evidence of macroeconomic effects of real exchange rates and exchange rate volatility is mixed, although overall, there are signs that REER depreciation does contribute to economic growth, particularly in developing countries ( Ridhwan et al., 2024 ). This research was concerned with the effect of exchange rate volatility on two macroeconomic trade relationships: bilateral trade and bilateral investment.

There have been several theories which have been used to address a potential relationship between exchange rate volatility and international trade flows. The classical theory of exchange rate hysteresis proposes that there is a feedback effect between exchange rates and trade flows, with firm entry through trade feeding back into exchange rates ( Baldwin, 1989 ). Misaligned exchange rate policies—for example, the use of a fixed exchange policy when a floating exchange policy would be better to achieve economic growth under the country’s current conditions have also been proposed as a theoretical explanation ( Baldwin & Lyons, 1994 ). It has also been proposed that the higher variability of the real price of goods under floating exchange rate policies could create short-run uncertainty ( Gali & Monacelli, 2005 ) and merchant risk aversion ( Viaene & de Vries, 1992 ), thus inhibiting trade. The dynamic equilibrium model ( Sercu & Uppal, 2003 ) has also been used to explain the feedback effect of exchange rate volatility into the complex macroeconomic system. However, none of these theories has been fully accepted, particularly given the mixed empirical evidence.

Previous studies have provided some empirical support for an influence of exchange rate volatility on international trade flows. One meta-analysis found generally negative effects of exchange rate volatility on trade flows, although these findings were very mixed ( Ozturk, 2006 ). These findings were supported by a later and more extensive meta-analysis ( Haile & Pugh, 2013 ), as well as a meta-regression ( Bouoiyour & Selmi, 2016 ). There have also been a few studies from East Asia and the ASEAN region, which have had mixed findings. One of these studies examined ASEAN-4 countries (Indonesia, Malaysia, Singapore, and Thailand), using data from each country’s five largest trade partners ( Soleymani et al., 2017 ). The authors found a short-run negative effect on both export and import flows. Another study, which examined Thailand and its 20 largest trading partners, found a negative long-run effect of exchange rate volatility, but no significant short-run effects on trade flows ( Dau & Sethapramote, 2021 ). A study of East Asian countries found a significant negative effect of exchange rate volatility on trade partners ( Hayakawa & Kimura, 2009 ). However, a study of ASEAN + 3 (including China, Japan, and South Korea) found no significant effect on bilateral trade flows between countries ( Nguyen & Vo, 2017 ), while a study of Malaysia-Thailand trade found variation at the industry level ( Aftab et al., 2017 ). A recent global study found that there was such an effect, but that crude oil prices were endogenous, meaning the effect disappeared once this was controlled ( Vilela Vieira & MacDonald, 2016 ).

In summary, the theoretical and empirical positions of the literature support a negative effect of exchange rate volatility on trade, although these studies are mixed in their methodologies, data sources, and findings. Thus, the research investigates the following:

Hypothesis 1: Higher rates of exchange rate volatility will have a negative effect on bilateral trade between ASEAN countries.

Data was collected for nine ASEAN countries (2000 to 2021). Myanmar was excluded from data collection due to inconsistent data availability. Data was sourced from a variety of standard databases, including the CEPII Gravity database ( CEPII, 2021 ; Conte et al., 2021 ), Refinitiv Eikon ( Refinitiv, 2022 ), the World Governance Indicators ( World Bank, 2021 ), and UNCTAD STAT ( UNCTAD, 2022 ). Additional data was added, including the FDI restrictiveness index and the normalized Chinn-Ito index of financial openness ( Chinn & Ito, 2023 ) to reflect regulatory pressures and constraints on trade and FDI. A summary of the variables and data sources is provided in Table 1 . It should be noted that the series for trade flows (TRADE) could only be sourced to 2020; therefore, the analysis period for trade is shorter than the analysis period for FDI flows.

The research used the gravity model as the basis for analysis. The gravity model is a theoretical model which assumes that economic exchange is higher between countries that are ‘closer’ geographically or politically, based on Heckscher-Ohlin and Ricardian theories of competitive advantage and competition ( Yotov et al., 2016 ). It assumes that elasticity of demand, substitutability, and costs of trade drive trade between two countries ( Yotov et al., 2016 ). Costs of trade depend on factors like geographic distance and adjacency (common borders), existing political linkages such as historic colonial relationships, and cultural ‘distance’ factors such as shared language ( Yotov et al., 2016 ). The basic form of the gravity model is stated as:

$\log X_{i,j}={\beta }_0+{\beta }_1\log GD{P}_{i-j}+{\beta }_2\log DIS{T}_{i,j}+{\beta }_3\log POP{D}_{i,j}+{\epsilon }_{i,j}$ (1)

In which X_i,j represents bilateral trade flows, GDP_i-j represents difference in GDP of trade partners, DIST_i,j represents geographic distance, and POPD_i,j represents population difference ( Shepherd, 2016 ). To this basic equation, other terms may also be added to represent common language, historic ties, shared borders, and other control variables ( Shepherd, 2016 ).

In this research, the standard form of the variables, rather than the log form, was used. This was due to empirical specification problems with using log transformations, in that the number of missing variables due to negative values significantly degraded the dataset. This research added an additional control variable of political instability (POLINST), due to high variations within the region and its acknowledged role in trade flows ( Broda & Romalis, 2011 ). While the gravity model specifies common official language as a potential factor in trade ( Anderson, 2011 ), in practice, there were few shared languages except between Singapore and Malaysia, and therefore, this was eliminated due to invariance. Instead, a variable representing the regulatory restrictiveness of the trade climate were added as controls. This variable was the Chinn-Ito index of financial openness (OPEN), which indicates regulatory constraints on capital movements ( Chinn & Ito, 2023 ). The Chinn-Ito index is a normalized coefficient, with 0 indicating complete capital closure and 1 indicating complete capital openness ( Chinn & Ito, 2023 ). Given the definitions of the variables, it is expected that OPEN will be related to both TRADE and FDIF.

There were two final equations constructed based on this model, investigating the effects of exchange rate volatility (EXCHVOL) on bilateral trade flows (TRADE) (Equation (2)) and FDIF (Equation (3)) respectively.

$\begin{array}{c}\log TRAD{E}_{i,j}={\beta }_0-{\beta }_{1}EXCHVO{L}_{i,j}+{\beta }_{2}GD{P}_{i-j}-{\beta }_{3}DIS{T}_{i,j}-{\beta }_{4}POP{D}_{i,j}\\ -{\beta }_{5}POLINS{T}_i+{\beta }_{6}OPEN+{\epsilon }_{i,j}\end{array}$ (2)

$\begin{array}{c}\log FDI{F}_{i,j}={\beta }_0-{\beta }_{1}EXCHVO{L}_{i,j}+{\beta }_{2}GD{P}_{i-j}-{\beta }_{3}DIS{T}_{i,j}-{\beta }_{4}POP{D}_{i,j}\\ -{\beta }_{5}POLINS{T}_i+{\beta }_{6}OPEN+{\epsilon }_{i,j}\end{array}$ (3)

Analysis

Following common practice with the gravity model ( Shepherd, 2016 ), analysis was conducted using a fixed effects panel regression approach. Analysis was conducted in the GRETL econometric software package.

Table 2 summarizes the analysis of Equation 2 (EXCHVOL → TRADE) in each of the nine ASEAN countries investigated. Autocorrelation was investigated using the Durbin-Watson (d) statistic, which in all cases indicated no excessive autocorrelation (p > 0.05). However, residuals were not normally distributed in most of the datasets (p < 0.001), and there was evidence for heteroskedasticity (Wald chi-square p < 0.001) in most datasets as well. Thus, there was some

Note. Dependent variable: TRADE; *p < 0.10, **p < 0.05, ***p < 0.01; . = Variable eliminated due to high collinearity

variance from the assumptions of panel analysis, which were likely due to the relatively small series lengths ( Wooldridge, 2013 ). Such violations can introduce bias into the estimates ( Wooldridge, 2013 ), which was considered as an issue, but as there was no way to adjust the length of the data series, the analysis continued.

The relationship of EXCHVOL and TRADE was the central issue of concern. There were a mixture of findings. With the exception of Philippines and Thailand, the EXCHVOL→TRADE was negative. However, the effect of EXCHVOL on TRADE was not significant in Laos, and in Malaysia it was marginally significant (p < 0.10). These findings indicate that overall, EXCHVOL has a negative relationship to TRADE, which was anticipated given the literature. However, also per prior studies, there was a mix of directionality, and the strength of the effect varied. This may be due to the influence of various control variables; for example, GDPPCD, DIST, POPD, and POLINST, which had significant effects in some countries with varying directionality, and OPEN, which was significant for all countries except Vietnam, but had varying directionality as well. Thus, H1 was supported to a great extent, with most countries exhibiting a negative effect of EXCHVOL on TRADE. However, this effect was not entirely consistent.

Table 3 summarizes the analysis of Equation 3 (EXCHVOL → FDIF) in nine ASEAN countries. As with the previous analysis, the Durbin-Watson statistic (p < 0.001) indicated no excessive autocorrelation, but residuals were not normally distributed in most datasets (p < 0.001) and there was evidence of heteroskedasticity according to the Wald test in many of the datasets as well (p < 0.001). Therefore, as expected these data series did violate some of core assumptions of regression, which was likely due to the small dataset size.

The central EXCHVOL→FDIF relationship was not entirely consistent. Brunei, Cambodia, and Malaysia exhibited a negative effect of EXCHVOL on FDIF, but the effect of EXCHVOL on FDIF in the Philippines was positive. Indonesia, Laos, Singapore, Thailand, and Vietnam did not show a significant effect of EXCHVOL on FDIF. Control variables also had variable effects within this model, raising the possibility that within-country variations in these factors could also have an effect on outcomes. In particular, OPEN had a significant positive effect in several countries, which indicates that the overall capital openness of the country has a significant influence on FDI. Overall, there was only weak support for Hypothesis 2, and it is clear that this relationship is dependent on exogenous factors.

Note. Dependent variable: FDIF; *p < 0.10, **p < 0.05, ***p < 0.01; . = variable eliminated due to collinearity.