The Dynamics of Three Currencies’ Bid–Ask Spread From the Perspective of Import–Export Success: Novel Findings for Digital Trade and Sustainable Production

Import and export have been acknowledged for their beneficial effects on sustainable economic development. In the context of economic globalization, the dynamics of exchange rates are more critical and necessary to export success and sustainable development. Therefore, understanding the dynamics of exchange rates contributes to the achievement of export success so as to promote sustainable production. The aim of this research was to find out the dynamics of spreads in foreign exchange rates over 15 years and the systematic relationship between dynamics of spreads in foreign exchange rates and information arrival and dynamics of spreads and cost of carry. The multiple regression, dummy variable test, and vector autoregressive model show that most variations in the spreads result from relative spread rather than information arrival and cost of carry for the long term. The information arrival and cost of carry have almost no influence upon relative spread, even during the global financial crisis in 2008. This article suggests that exporters should pay more attention to the effects of relative spread rather than information arrival and cost of carry in the foreign exchange market when they carry out foreign trade.


INTRODUCTION
Sustainable development answers to the common needs and demands of human beings in both developed countries and developing countries . It is widely accepted that economic growth cannot be achieved at the cost of the environment; thus, sustainable development has become a common aspiration of the society . Studies have shown that sustainable economic development can be achieved by using cleaner energy, improving the efficiency of using natural resources, green technology innovation, and environmental regulations Song et al., 2022a;Song et al., 2022b;Zhao et al., 2022c). Moreover, some studies have shown that there is a positive relationship between government expenditure and economic sustainable growth (Delani et al., 2021), and policymakers should promote green finance to provide more funds for green energy for sustainable economic development (Samour et al., 2022), but they need to pay attention to the effects of external shocks such as oil prices on domestic markets to maintain economic stability (Alhodiry et al., 2021). Furthermore, import and export have been acknowledged for their beneficial effects on sustainable economic development (Zhao et al., 2022d;Magazzino et al., 2022;Shahzad et al., 2022). Magazzino et al. (2022) and Shahzad et al. (2022) point out that sustainable economic growth is found to be substantially driven by trade globalization diversification. In the context of economic globalization, the dynamics of exchange rates are more critical and necessary to export success and sustainable development. Therefore, understanding the dynamics of exchange rates contributes to the achievement of export success so as to promote sustainable production.
Both developed and developing countries would like to explore the international market because of export success and sustainable economic development, and thus they have to be concerned about foreign exchange trading. The bid-ask spread in foreign exchange markets is dealers' profit by trading execution. Some researchers found that costs have a relationship with the bid-ask spread (Demsetz, 1968;Benston and Hagerman, 1974). Bessembinder (1994) argued that foreign exchange dealers adjust the spread to react to changing expectations about the cost of carry. Boothe (1988) and Glassman (1987) highlighted volatility of having a positive relationship with spread. In addition, Galati (2000) found that volatility of emerging market currencies caused by the public information arrival has a positive relationship with spread.
The analysis of changes in the cost of carry and information arrival is mainly concerned for observing spread variations for dealers' trading (Chelley-Steeley and Tsorakidis, 2013). However, such analysis has been rare, especially one focusing on the two factors during and after the global financial crisis, which may draw many countries' attention. This research investigates how changes in the cost of carry and information arrival affect the evolution of the bid-ask spread and whether these two factors changed during the global financial crisis in 2008. It is expected to fill the research gap.
This research work applies the three main variables-volatility of information arrival, cost of carry, and relative spreads, which are similar to those used by Chelley-Steeley and Tsorakidis (2013). We tend to perform a time-series model, for which stationary of data should be determined first. Hence, we use the ADF test to make sure whether the data collected are stationary or not. We find that there are more than one nonstationary data, so we perform the cointegration test. After the cointegration test, we can verify the cointegration relationship between a set of variables (dependent and explanatory variables) which is equivalent to testing the stationarity of the residual series of the regression equation. Next, we conduct a multiple regression. When we perform the Chow test and dummy variable test, we add a dummy variable-the global financial crisis-from the dummy variable test model. We consider that the global financial crisis is an important factor and tend to know whether the crisis can affect the relationship among the information arrival, cost of carry, and relative spread, so we choose the crisis as a dummy variable. We also use a vector autoregressive model (VAR) test, which is similar to the methodology applied by Chelley-Steeley and Tsorakidis (2013).
The results show that information arrival and cost of carry have almost no impact on the spread variation, and most variations are attributed to the spread itself. Moreover, it is found that the global financial crisis in 2008 did not have a significant impact on the spread, information arrival, or cost of carry. It can be a reference for both developed and developing economies that focus on foreign exchange markets.
The rest of this article is organized as follows: Section 2 presents the literature review. Section 3 describes data and summary statistics. Section 4 illustrates the methodologies. Section 5 presents the results of the research. Section 6 is the conclusion.

LITERATURE REVIEW
It is necessary to know the situation of bid-ask spread of some main foreign currencies for most of the countries, so they can apply these currencies more efficiently. Some economists have conducted research on foreign exchange trading. Some of them found the relationship between cost components and trading over time. Demsetz (1968) found that changes in costs (such as order processing cost and cost of carry) usually have a significant effect on the bid-ask spread. In addition, Benston and Hagerman (1974) pointed out that order processing costs play an important role in the evolution of bid-ask spread. Bessembinder (1994) argued that foreign exchange dealers adjust the spread to react to changing expectations about the cost of carry. Galati (2000) found that volatility of emerging market currencies caused by the public information arrival has a positive relationship with effective spread. Two researchers have verified that the association between volatility and the bid-ask spread is positive. Boothe (1988) argued that the spread is mainly determined by exchange rate uncertainty after comparing the data of seven countries. Glassman (1987) showed that exchange rate risk, modeled by the lagged absolute change, is also a main determinant of the exchange rate. In addition, some researchers have examined liquidity in the foreign exchange market. For instance, Froot and Ramadorai (2005) and Breedon and Vitale (2010) indicated that the understanding of how liquidity affects exchange rates and the spread is insufficient. Moreover, Evans M. and Lyons R. (2002) showed that liquidity of the DM/US dollar does not relate to return volatility and trading volume. For the case of G7 and E7 economies, Rafique et al. (2021) documented that economic complexity as product diversification of export basket has direct linkages with greener energy generation, which contributes to sustainable production. More recently, Magazzino et al. (2022) reported that export product diversification is directly linked with energy use and sustainable production in APEC countries.
Two studies examined currency trading and quotation, but both of them focused on particular foreign exchange markets. Hua (2009) pointed out that large multinational banks expand spread quotation to respond to the quotation widened by small Taiwanese banks in the Taipei inter-bank foreign exchange market. Cheung and Chinn (2001) showed that individual macroeconomic variables significantly influence exchange rate dynamics in the United States foreign exchange market.
Our research is based on the study by Chelley-Steeley and Tsorakidis (2013), which shows that changes in inventory, adverse selection, and cost of carry have little influence on further spread variation. This means that the three variables (volatility, quote revision, and interest rate differentials) have a much smaller impact on the spread than the past or future spread itself. They used foreign exchange quotes of ten years (from 1995 to 2005), and they only analyzed the long-term period data without considering important information events. In our research, updated data from 2001 to 2016 were selected. This period is not only over ten years but also contains an important information event. Moreover, our study separated the long period into three short periods to determine whether some information events significantly impact bid-ask spreads.

DATA AND SUMMARY STATISTICS
US dollar/UK pound and US dollar/Euro taken from the OANDA History database were used as spot exchange rate quotes in this study. The two quotes were selected because the US dollar is the main currency for international trade and investment, and the two currency pairs could be the most active currencies of trading in the foreign exchange market. The selected data are daily closing bid and ask quotes, and the monthly average was computed as monthly bid and ask quotes (Chelley-Steeley and Tsorakidis, 2013). Indicative quotes rather than transaction prices were chosen because they contain more information (Goodhart and O'Hara, 1997;Phylaktis and Chen, 2009).
The closing relative spread for each monthly closing quote was calculated using a method similar to that used by Ding and Charoenwong (2003) and Wang (2001). The calculation formula is as follows: Relative spread% (ask − bid) (ask + bid)/2 where ask is the closing ask quote and bid is the closing bid quote. Figure 1 shows the monthly average of the daily spread of the US dollar/UK pound from 1 January 2001 to 30 September 2016. Figure 2 shows the same daily spread for the US dollar/Euro. It can be seen that the spreads of the US dollar/UK pound and the US dollar/Euro decreased from 2001 to 2007. Between 2008 and 2010, when the global financial crisis broke out, spreads of both US dollar/UK pound and US dollar/Euro fluctuated, and those of the US dollar/Euro fluctuated more gently. From 2010 to 2016, the spreads of the two currency pairs were more stable than in previous periods and that of the US dollar/UK pound fluctuated more significantly.
The effects of volatility of information arrival and cost of carry have also been shown to be significant. In the aspect of the studies on volatility, Galati (2000) found that volatility caused by the public information arrival and that of currencies of emerging markets have a positive relationship with effective spread. Huang and Masulis (1999) found that volatility is positively related to the bid-ask spread in the foreign exchange market. Moreover, Evans M. D. D. and Lyons R. K. (2002) pointed out that order flow is highly correlated with exchange rate dynamics as it can convey fundamental information. In addition, Melvin and Yin (2000) argued that exchange rate volatility is linked to information arrival.
Many foreign exchange volatility metrics have been proposed in previous research. Some researchers used variables such as price volatility of foreign exchange future contracts or GARCH (Generalized Autoregressive Conditional Heteroskedasticity) model to estimate the conditional variance of the spot exchange rate (Glassman, 1987;Boothe, 1988;Bessembinder, 1994;Bollerslev and Melvin, 1994). Moreover, Huang and Masulis (1999) used the variance of bid-ask midpoint quotes within 15 min to measure volatility. Mclnish and Wood (1992) used the standard deviation of relative spread to measure volatility. Kim and Rhee (1997) utilized [ln (p t /p t-1 )] 2 to capture price volatility. In this formula, p t is the closing price at the t time. Our research captured the volatility of prices using the variation coefficient (C) of the mid-prices within each trading day or time interval. The calculating method is the same as the one used by Ding and Charoenwong (2003). The formula is as follows: where SD is the standard deviation of the mid-prices.
In the aspect of the studies on the cost of carry, Bessembinder (1994) pointed out that the interest-rate-based measure of liquidity costs is helpful for the measure of the cost of carry and argued that "the opportunity cost resulting from the requirement to maintain liquid inventories is the difference between the interest earned on highly liquid positions and the interest that could have been earned on less liquid positions" (Bessembinder, 1994, p. 323). He showed one measure of liquidity costs using the following formula: Cost of carry one − month Eurodollar deposit rate − overnight Eurodollar deposit rate. (2013) also used this formula, and it was applied in our research. In order to capture the cost of carry, the data of the 1-month Eurodollar daily deposit rate and Overnight London Interbank Offered Rate (LIBOR; based on US dollar) were collected from the economic research FRED economic database. The period for those data is also from 2001 to 2016.

METHODOLOGY
Previous studies have utilized questionnaire surveys. For example, Hua (2009) sent questionnaires to 281 recipients, including 110 financial staff at 36 foreign banks and 171 financial staff at 40 domestic banks. In addition, Cheung and Chinn (2001) used data obtained from foreign exchange traders by mail surveys in the United States. Nevertheless, a questionnaire survey is not a suitable methodology for our research. One important reason is that the response rate of questionnaires can be very low, which may cause inaccurate results. For instance, one mail survey on the US foreign exchange market had a response rate of approximately 8.1% (Cheung & Chinn, 2001). The other reason is that most questionnaire surveys are time-consuming as the designers need to spend a significant amount of time waiting for responses. Given that it is theoretically uncertain which methodology is superior, the exhaustive time-series analysis methodologies were chosen. This approach has the advantage of illustrating the differences among these methods on a common data set.

ADF Test
The ADF (Augmented Dickey-Fuller) test is used to determine whether the data in this research are stationary. There are two reasons to explain why this test is important. First, the stationarity of data can significantly affect their behaviors and properties. Second, the use of non-stationary data usually causes spurious regression; in other words, if non-stationary data are applied to the standard regression techniques, there will be a good but valueless regression in a standard measure (significant coefficient estimates and a high R 2 ). Therefore, non-stationary and stationary variables should be treated differently. In the ADF test, test equations with the constant and constant and trend were chosen.
The equation with the constant: The equation with the constant and trend: The hypothesis: Decision rule: If tp > ADF critical value, = => null hypothesis is not rejected, i.e., unit root exists.
If tp < ADF critical value, = =>null hypothesis is rejected, i.e., unit root does not exist.
The number of augmenting lags is determined by minimizing the Akaike information criterion or dropping lags until the last lag is statistically significant.

Cointegration Test
Regression analysis in econometrics requires stationary time series since non-stationary time series degrades the effectiveness of the analysis. To solve non-stationary time series found by the ADF test, Engle and Grander developed cointegration at the beginning of the 1980s. Cointegration implies that the linear combination of two or more nonstationary time series may contribute to stationarity. Cointegration focuses on the equilibrium relationship among more than two non-stationary time series. It is significant for establishing an econometric model with non-stationary economic variables and verifying the long-run equilibrium among these variables.
then Y is cointegrated, and β is called the cointegration vector. It should be noted that: 1) The cointegration vector, which describes the relationship among non-stationary variables, is not unique.
2) The order of integration of the cointegration vectors must be the same. 3) There may be as many as k-1 cointegration linearly independent vectors.
4) The cointegration vectors share the same stochastic trends and are proportional.
Engle and Granger proposed the E-G two-step method in 1987 to verify the cointegration relationship. If the linear combination of independent variables can explain the dependent variable, they have an equilibrium relationship. The part of the dependent variables that cannot be explained by the independent variable constitutes a stationary residual series. Therefore, verifying the cointegration relationship between a set of variables (dependent and explanatory variables) is equivalent to testing the stationarity of the residual series of the regression equation. The procedure of the E-G two-step test is as follows: 1) To establish a regression equation, if y 2t , y 3t , ... y kt , are all I(d): The estimated residual series iŝ 2) Next is to verify whether the residual seriesμ t is stationary with the ADF test, in other words, to determine whether it has a unit root. The critical value needs to be determined by referring to the

Multiple Regression
In this research, multiple regression analysis is used to discuss whether volatility and cost of carry influence bid-ask spread. Regression describes and evaluates the relationship between a given variable and one or more other variables. More specifically, regression explains variations in one variable by reference to variations in one or more other variables. The model is: where the variables X 2t and X 3t are explanatory variables that are thought to influence Y t , and the coefficient estimates β1, β2, and β3 are the parameters that quantify the effect of these explanatory variables on Y t .

Chow Test and Dummy Variable Test
The fourth methodology is the Chow test and the dummy variable test since a dummy variable is added. The dummy variable is related to the data span of the global financial crisis from 2008 to 2010. We consider the global financial crisis as an important factor. So, it is hypothesized that the global financial crisis affects the relationship between bid-ask spread and volatility and between bid-ask spread and cost of carry. In this research, the years 2008 and 2010 are selected to divide the sample period because the global financial crisis first occurred in 2008 and was widely thought to be finished by 2010. The Chow test model is: Frontiers in Environmental Science | www.frontiersin.org May 2022 | Volume 10 | Article 900760 5 There are three samples from different economic environments (before 2008, between 2008 and 2010, and after 2010), and their quantities are n 1 (<n), n 2 (<n), and n 3 (<n) with n 1 +n 2 +n 3 = n. The three samples are: Y 1t α 1 + α 2 X 2t + α 3 X 3t + ...α k X kt + ξ 1t , (t 1, 2, ...n 1 ); Y 1t β 1 + β 2 X 2t + β 3 X 3t + ...β k X kt + ξ 2t , (t n 1 + 1, n 1 + 2, ...n 2 ); Y 1t ε 1 + ε 2 X 2t + ε 3 X 3t + ...ε k X kt + ξ 3t , (t n 2 + 1, n 2 + 2, ...n).
The Chow test presents that the structure of regression is changed but does not show how the structure of regression changes. Therefore, the dummy variable test is applied. The dummy variables are: If the D t1 0, t > 2008; D t2 0, t > 2010; 1, t < 2008; 1, t < 2010.
For D t1 , 0 represents the period after 2008, and 1 represents the period before 2008; for D t2 , 0 stands for the period after 2010, and 1 stands for the period before 2010.
The dummy variable test model is: where β 1 to β 9 are parameters quantifying the effects of these explanatory variables on Y t ; β 4 D t1 and β 5 D t2 are addition forms of the dummy variable test; X 2t D t1 , X 3t D t1 , X 2t D t2 , and X 3t D t2 show multiplication forms of the dummy variable test.

Vector Autoregressive Model
The last methodology is the vector autoregressive model (VAR) test. This methodology is similar to the method used by Chelley-Steeley and Tsorakidis (2013). VAR methodology is an n-equation and n-variable model in which each variable is, in turn, explained by its own lagged value and current and past values of the remaining n-1 variables. For example, in a twovariable case, it is a model in which the time path of {Y t } is affected by current and past realizations and the time path of the {Z t } sequence and current and past realizations of the {Y t } sequence. The simple bivariate system is: In general, a VAR(p)-process would be: where p is the lag length, c is a constant vector, e t is the white noise, and Y t is a vector containing the time series. In the VAR system, impulse responses and variance decomposition are the most important. Impulse responses are useful to separate the impacts of different variables on spread evolution. In addition, in this research, impulse responses present how much a 1% shock from one variable influences other variables in VAR. "Moreover, the variance decomposition of the VAR will identify how much of the variation in the spread can be explained by each different component used in the VAR model" (Chelley-Steeley and Tsorakidis, 2013, p. 125). It determines whether changes in the volatility and cost of carry can influence unexpected changes in the relative spread. The lower value of the variance decomposition of one variable may indicate that other components are more important and have more impact on spread variation. Table 1 shows the result of the ADF test for the US dollar/UK pound in which relative spread, volatility, and cost of carry are represented by gbprs, gbpv, and gbpcoc, respectively, and are represented by dgbprs, dgbpv, and dgbpcoc, respectively, after the first-order differencing. Table 2 shows the same information for US dollar/Euro in which relative spread, volatility, and cost of carry are represented by eurrs, eurv, and eurcoc, respectively, and are represented by deurrs, deurv, and deurcoc after the first-order differencing.

ADF Test
The results of the ADF test before the differencing of the time series are shown in Table 1 and Table 2. It can be seen that the ADF values of time series are greater than the critical values of 1 and 5%, and the p-values are greater than 0.05. Therefore, the null hypothesis that the time series is non-stationary should be accepted. However, after first-order differencing, the ADF values are less than the critical value of 1 and 5%, and the p-values are less than 0.05. Therefore, the null hypothesis is rejected, and the time series after first-order differencing is regarded as stationary. In addition, after first-order differencing, all of the time series are I(d). The prerequisite of the cointegration test is realized.

Cointegration Test
It can be concluded from the previous ADF test that the time series used in this study are all I(d), which guarantees the feasibility of cointegration analysis. For the cointegration analysis of the relationship between the relative spread and the other two variables, the three variables involved in US dollar/UK pound and US dollar/Euro were regressed with ordinary least squares (OLS), and the regression equations were established. Then, the ADF test of the residual series of the equations was conducted, and the results are shown in Table 3 and Table 4.
It can be seen that the ADF values of the residual series of the cointegration equations of US dollar/UK pound and US dollar/ Euro are -2.6126 and -2.6016, respectively, significantly lower than the critical value of -2.5776 under the confidence level of 0.01. In addition, p-values are close to zero. Therefore, the hypothesis that unit root exists is rejected, and the residual series is considered stationary. Accordingly, it can be determined that the cointegration relationship of gbprs, gbpv, and gbpcoc exists, and that of eurrs, eurv, and eurcoc also exists. In other words, there is a stable long-run equilibrium among the variables.

Multiple Regression
The relationships between the volatility of information arrival and relative spread and between the cost of carry and relative spread of the two currency pairs are shown in Table 5. It can be seen that there is almost no relation between the two independent variables and the relative spread. For the US dollar/ UK pound, the coefficients are both small and close to zero (-0.004627 for volatility and 0.000032 for the cost of carry), suggesting that changes of volatility and cost of carry have a small impact on the bid-ask spread, although the p-values of volatility and cost of carry are less than 0.05. In addition, the R-squared is 0.16, which means only 16% of data can be fit into this regression model, demonstrating that there are hardly any relationships between two independent variables and relative spread for the US dollar/UK pound. Similar phenomena were observed for the US dollar/Euro. The coefficients of the two variables are approximately zero though p-values are more than 0.05 (0.240 and 0.079, respectively), indicating that the two independent variables have scarcely any relationship with the relative spread of US dollar/Euro. These results are different from those in previous research. Most previous studies showed that volatility and cost of carry impact relative spread (Benston and Hagerman, 1974;Bessembinder, 1994). For example, Huang and Masulis (1999) pointed out that volatility is positively related to the relative spread in the foreign exchange market; Stoll (1978)

Chow Test and Dummy Variable Test
The Chow test results of the two currency pairs are illustrated in Table 6. It can be seen that the F-statistic of US dollar/UK pound and US dollar/Euro have been low since the year 2008, and the year 2010 was chosen to divide the sample period. In addition, the p-values of both two currency pairs are over 0.1, indicating that regressions were not different in the three different periods.  The dummy variable test shows more details about the differences in the regression in the three periods. The results of the Dummy variable tests of US dollar/UK pound and US dollar/Euro are shown in Table 7, where D t1 and D t2 represent dummy variables for 2008 and 2010. The global financial crisis hardly affects the relative spread of the two currency pairs, as demonstrated by that t-Statistics of the two dummy variables in 2008 and 2010 are not significant, and p-values are more than 0.1. The coefficients of the two dummy variables are all small (3.88E-06 and -5.86E-06 for US dollar/UK pound, -7.12E-06 and 2.11E-07 for US dollar/Euro). In addition, the global financial crisis does not significantly impact volatility and cost of carry, as shown by low t-statistics and high p-value (more than 0.1). The R-squared values are 0.17 and 0.04 for two currency pairs, respectively, which means only 17 and 4% of data can be fit into those regression models. These results indicate that the regressions in the three different periods were not different.
It was hypothesized that the global financial crisis in 2008 would impact volatility, cost of carry, and spread. However, the results of our research show that there were no significant changes in regressions in the three different periods, indicating the global financial crisis in 2008 had little impact on the three variables of the two currency pairs.

Vector Autoregressive Model
It can be seen from the results of the previous ADF test that the six time series variables of US dollar/UK pound and US dollar/Euro are non-stationary and become stationary after first-order differencing. Therefore, the data after first-order differencing are used to establish the VAR model. According to the principle of using the lowest Akaike information criterion and Schwartz criterion, the lag order of the variables is determined to be first-order after several attempts. The estimation results of the regression are as follows: US dollar/UK pound  It is also necessary to test the stability of the estimated model. If the established VAR model is not stable, some results will be invalid (e.g., the standard error of the impulse response function). Therefore, this research applies the AR root for verification. If the  modulus of all roots is smaller than 1 (i.e., the unit roots lie inside the unit circle), the estimated model is stable and otherwise unstable. The results are shown in Figure 3.
It can be seen from Figure 3 that all unit roots of the estimated VAR models lie inside the unit circle, indicating that the models are stable. Therefore, impulse response analysis and variance decomposition analysis on the two models can be performed.

Impulse Responses
The results of the standardized impulse responses are shown in Figure 4 and Figure 5. Figure 4 presents the results of the US dollar/UK pound. Figure 5 shows the same information for the US dollar/Euro. The impulse responses explain how much impact a 1% shock from one variable has on the other variables. Simply speaking, they indicate how shocks from each variable are transmitted to the spread. Figure 4 (response of dgbprs to dgbpv) illustrates that the relative spread shows a small negative response after a positive shock is given to volatility in the first period. It reaches the maximum in the second period and converges. It is indicated that volatility has a lagged effect on the spread in the short run; however, the change of volatility does not have a significant driving effect on the spread. The negative effect on spread tends to zero after the second period and then levels off. It can be concluded that a positive shock on volatility has a small negative effect on the relative spread in the short run, but the long-run effect tends to be zero. Overall, there is no significant driving effect of volatility in the US dollar/UK pound on the spread. Figure 4 (response of dgbprs to dgbpv) also suggests that the relative spread shows a trend of small positive response after a positive shock is given to the cost of carry in the first period. It reaches the maximum in the second period and converges. It is indicated that the cost of carry has a lagged effect on spread in the short run; however, the change in the cost of carry does not have a significant driving effect on the spread. The positive effect on spread tends to be zero after the second period and then level offs. It can be concluded that a positive shock to volatility has a small positive effect on the relative spread in the short run, but the longrun effect tends to be zero. Overall, there is no significant driving effect of the cost of carry in US dollar/UK pound on the spread.
In addition, in US dollar/UK pound, the spread is significantly influenced by its own shock in the first period (Response of dgbprs to dgbprs), but the impact converges and tends to zero after the second period, indicating that the impact on the relative spread is largely attributed to the short-term shock of the spread itself.
The impulse response of the US dollar/Euro shows the same results. As shown in Figure 5, the spread is significantly impacted by its own shock in the first period (response of deurrs to deurrs), but the effect converges and tends to zero after the second period, indicating that the effect on the relative spread is largely attributed to the short-term shock of spread itself. Shocks on volatility and cost of carry have almost no effect on the spread.

Variance Decomposition
The results of the variance decomposition are summarized in Table 8 and Table 9. Table 8 displays the results for the US dollar/UK pound, while Table 9 shows the results for the US dollar/Euro. Variance decomposition evaluates the importance of different structural shocks by analyzing the degree of contribution of each structural shock to the changes in endogenous variables (usually measured by variance). The variance decomposition provides the relative importance of each random disturbance affecting the variable in the VAR model. The response lag and relative effect of the variable can be estimated by comparing the change of relative importance over time. The variance decomposition results after the 10th period, which stabilize, are not presented in this article. It can be seen from Table 8 that the variation of the growth rate of the relative spread of the US dollar/UK pound is mainly attributed to the shock itself. The growth rate reaches 100% in the first period and then tends to decrease. The decrease is not significant, and the growth rate stabilizes at 94% after the 7th period. The contribution of the shocks from the variation of the change rate of volatility to the fluctuation of the growth rate of spread only accounts for 3% in the second period. In other words, 3% of the forecast variance of the growth rate of the spread can be explained by the variation of the change rate of the volatility. It is indicated that the effect of the change rate variation of the volatility on the growth rate of the spread is small, and this effect changes slightly over time. The contribution maintains at about 3% in the end. It can be seen from the aforementioned analysis that although the variation in change rate of volatility has an impact on the variation of the growth rate of relative spread, the variation in the growth rate of relative spread is mainly attributed to the change in itself.
It can also be found from Table 8 that the contribution of the shock from the variation of the change rate of cost of carry to the fluctuation of the relative spread (i.e., the contribution to forecast error) accounts for 2% of the total forecast errors in the second period. The relative contribution maintains at about 2% afterward. It is indicated that the change rate variation of the cost of carry only explains 2% of the variation of the growth rate of the spread. It can be concluded that the influence of the change rate of the cost carry on the growth rate of the spread, whether in the short or long run, is small. This result is consistent with that of impulse response analysis. The variance decomposition of the relative spread of the US dollar/Euro shows a similar result. It can be found from Table 9 that the change in the growth rate of the relative spread is mainly attributed to the shock from itself. The growth rate reaches 100% in the first period and tends to decline. The decline is insignificant, and the growth rate maintains at about 97% after the seventh period. The contribution of the shock from the change rate of volatility to the fluctuation of the growth rate of spread only accounts for 2% in the second period. In other words, only 2% of the forecast variance can be explained by the variation of the change rate of the volatility, indicating that the influence of the change rate of volatility on the spread is very small. The effect varies slightly with time, and the contribution maintains at about 2% in the last period. It can be seen from the aforementioned analysis that although the variation in the change rate of volatility has an impact on the variation of the growth rate of relative spread, the variation in the growth rate of relative spread is mainly attributed to the change in itself.
Table 9 also shows that the contribution of the shock from the variation of cost of carry on the fluctuation of the growth rate of relative spread (i.e., the contribution to forecast error) accounts for 2% of the total forecast errors in the second period. After that, the relative contribution maintains at about 0.1%, indicating that the change rate variation of the cost of carry has almost no influence on the growth rate of the spread, and it only explains 2% of the change of the growth rate of the spread. This result is consistent with that of impulse response analysis.
We mentioned five different methodologies previously. First, we find that the six time series variables of US dollar/UK pound and US dollar/Euro are non-stationary and become stationary after first-order differencing and use the ADF test to make sure the feasibility of cointegration analysis. Second, the result of the cointegration test shows that there is a stable long-run equilibrium among the variables. Third, using multiple regression, we find that there is almost no relation between the volatility of information arrival and relative spread and between the cost of carry and relative spread of the two currency pairs. Fourth, the results illustrate that there were no significant changes in regressions in the three different periods, indicating that the global financial crisis in 2008 had little impact on the three variables of the two currency pairs, through the Chow test and dummy variable test. Finally, we find that the fluctuation of relative spread is from itself, not from the fluctuation of arrival information and cost of carry, through impulse responses and variance decomposition.

CONCLUSION
This research studied the dynamics of the bid-ask spread, which is useful for economic globalization and sustainable developing research. The relative spread of two currency pairs commonly concerned by most of the countries, US dollar/UK pound and US dollar/Euro, were studied with a new data set that includes daily data over 16 years. This research focused on how changes in the components of the spread influence the relative spread's future dynamics and whether the influencing mechanism changed during different periods (especially during the important event), which could be a reference for both developed and developing economies that focus on foreign exchange markets.
The results show that the changes in the arrival of new information and cost of carry have almost no impact on future variations in the spread, even during the global financial crisis in 2008. The spread's future value is mainly determined by its dynamics, which could be beneficial for those who are concerned about foreign exchange markets. In the future, the influence of other factors (such as trading volumes and market frictions) on spread variation should be studied, which were not analyzed in this study because data on them were unavailable.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

LX:
writing-original draft and supervision. XL: conceptualization, methodology, and writing-review and editing. LS: project administration and software. WS: resources and funding acquisition. XZ: formal analysis, investigation, and validation.