This part uses Eviews9.0 to conduct an empirical analysis of the bilateral trade model. Before the overall regression, the article first conducts unit root test and cointegration test on the relevant panel data of 45 countries along the “Belt and Road” to avoid data non-stationary and pseudo-regression. After passing Hausman's test, the article uses random effects to perform regression analysis on the overall bilateral trade model. In order to further illustrate the degree of influence of various factors on trading countries in different regions, the article will divide the 45 countries along the route into 5 groups of Central Asia, South Asia, Central and Eastern Europe, ASEAN and West Asia for robustness testing.
4.1 The Empirical process of the overall bilateral trade model
As the article selected relevant data from 2008 to 2018, involving three dimensions of time, country and variables, it is suitable to use a static panel model for analysis. Pseudo-regression is prone to appear in the panel model, so the article chooses four methods of unit root test to measure the stability of the data to avoid this problem. It can be seen from Table 2 that the original sequence of "LNWEF" is stationary data, while the original series of "LNTRA?", "LNGDPi?" and "LNGDPj?" are non-stationary data, indicating that there is a unit root, and it is necessary to make a difference for the relevant variables and continue to test them.
Table 2 Unit root test of each explanatory variable
|
ADF-FisherTest
|
LLCTest
|
IPSTest
|
PP-FisherTest
|
LNTRA
|
191.907
(0.00)
|
-19.09
(0.00)
|
-4.46
(0.00)
|
91.444
(0.4377)
|
LNGDPi
|
163.938
(0.00)
|
-20.10
(0.00)
|
-3.01
(0.00)
|
9.680
(1.00)
|
LNGDPj
|
140.698
(0.0005)
|
-7.56
(0.00)
|
-1.96
(0.026)
|
116.539
(0.0314)
|
LNWEF
|
165.131
(0.00)
|
-19.35
(0.00)
|
-4.24
(0.00)
|
218.508
(0.00)
|
Note: the data in brackets are the concomitant probability of the statistics, and the data above the brackets are t statistics.
The first-order difference series of "LNTRA?" is stationary, while the first-order difference series of "LNGDPi?" and "LNGDPj?" are still non-stationary data. Therefore, these two variables need to be further tested for the second-order difference series. The results in Table 3 show that the quadratic difference sequence of "LNGDPi?" and "LNGDPj?" is stable.
Table 3: Unit root test for non-stationary explanatory variables
|
ADF-FisherTest
|
LLCTest
|
IPSTest
|
PP-FisherTest
|
DLNTRA
|
191.907
(0.00)
|
-19.09
(0.00)
|
-4.46
(0.00)
|
-4.021
(0.00)
|
DLNGDPi
|
163.938
(0.00)
|
-20.10
(0.00)
|
-3.01
(0.00)
|
68.312
(0.957)
|
DDLNGDPi
|
163.938
(0.00)
|
-20.10
(0.00)
|
-3.01
(0.00)
|
183.234
(0.00)
|
DLNGDPj
|
162.667
(0.00)
|
-7.56
(0.00)
|
-2.541
(0.055)
|
259.867
(0.00)
|
DDLNGDPj
|
162.667
(0.00)
|
-7.56
(0.00)
|
-4.360
(0.00)
|
259.867
(0.00)
|
Note: Δ is the first-order difference sequence, and △ is the second-order difference sequence. The data in brackets is the concomitant probability of the statistic, and the data above the bracket is the t statistic.
From the above analysis, we can see that there is a unit root for the "LNTRA?" sequence, and two-unit roots for the "LNGDPi?" and "LNGDPj?" sequences, that is, the variable data of the bilateral trade model is not stationary. According to the above analysis, regression analysis cannot be carried out directly at this time. For this non-stationary panel variable, the authors chose the ADF cointegration test. The results are shown in Table 4. Whether it is the ADF cointegration test within or between dimensions, the corresponding p values are all rejected at the 1% significance level of the null hypothesis, which indicates that there is a co-integration relationship between the panel data of the bilateral trade model at this time, and the phenomenon of pseudo regression can be avoided when performing regression analysis.
Table 4: ADF cointegration test
Explained variable
|
explanatory variable
|
|
t statistics
|
probability value p
|
LNTRA
|
LNGDPi、LNGDPj、LNWEF
|
In-group statistics
|
-4.222998
|
0.0000
|
In-group statistics
|
-6.419179
|
0.0000
|
The article uses the statistical test method of the Hausman test to determine the appropriate measurement method for the bilateral trade model. According to the principle of the test, if the p value of the test result is far more than 0.05, the null hypothesis is accepted and choose random effect, otherwise the fixed effect is selected. The test results are shown in Table 5. The p value of the Hausman test result is 0.1221, indicating that the bilateral trade model uses random effects to measure the effect better.
Table 5: Hausman test
Testing method
|
Chi-Sq.Statistic
|
Chi-Sq.d.f.
|
Prob
|
Random cross section
|
7.273261
|
4
|
0.1221
|
After dealing with the non-stationarity of the panel data and determining the random effects of the model, this paper uses Eviews 9.0 to conduct a multiple regression analysis on the bilateral trade model, and obtains the degree of influence of each variable on the bilateral trade volume from 2008 to 2018.In order to strengthen the explanation of GDP, geographical distance, port infrastructure quality and other influencing factors, this paper adopts the method of gradually introducing regression variables. The empirical results are shown in Table 6 model (1) ~ (5).
We can see from Table 6 that the modified R2 of model (5) is 0.58, the F statistic is 97.24, and the p value is 0.00. It can be seen that the overall goodness of fit of the bilateral trade model is relatively good, and it is significant at the level of1%. The significance level of LNGDPi and LNGDPj in each model exceeds 1%. The regression coefficients of LNGDPi and LNGDPj were 0.571, 0.567, 0.561, 0.542, 0.542 and 0.724, 0.707, 0.718, 0.766, 0.764, respectively. On the whole, for every 1% increase in GDPi, the bilateral trade volume will increase by approximately 0.54%, and for every 1% increase in GDPj, the bilateral trade volume will increase by approximately 0.76%. Therefore, the economic scale of China and the countries along the “Belt and Road” can significantly promote bilateral trade. The growth of GDP, and the driving force of the GDP of trading countries is greater. The sign of the LNDIS regression coefficient is sometimes positive and sometimes negative, but from the perspective of model (5), the regression coefficient is -1.115 and the p-value is 0.06, indicating that for every 1% increase in geographic distance, the bilateral trade volume decreases by approximately 1.115%, and Significant at the 10% level, which is consistent with the expectations. The regression coefficients of LNWEF are all positive. On the whole, for every 1% increase in the port infrastructure quality index, the bilateral trade volume will increase by approximately 0.33%, which is significant at the 5% level.
The three dummy variables BOR, FTA and APEC can all have a positive effect on the increase of bilateral trade volume, and FTA has a higher degree of influence. From the perspective of model (5), for every 1% increase in BOR, the bilateral trade volume will increase by approximately 0.23%; for every 1% increase in FTA, the bilateral trade volume will increase by approximately 1.67%, which is significant at the level of 1%; for every 1% increase in APEC, The bilateral trade volume increased by approximately 0.04%.
Table 6:Regression results of bilateral trade model
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
LNGDPi
|
0.570869***
(12.95008)
|
0.566665***
(12.86719)
|
0.561290***
(12.73171)
|
0.541612***
(12.50870)
|
0.542468***
(12.46479)
|
LNGDPj
|
0.723938***
(0.060473)
|
0.706809***
(11.51611)
|
0.718168***
(11.70002)
|
0.766241***
(13.46972)
|
0.763891***
(13.15731)
|
LNDIS
|
0.088173
(0.216167)
|
0.007831
(0.019096)
|
0.555110
(1.065979)
|
-1.126008**
(-1.972980)
|
-1.114786*
(-1.884687)
|
LNWEF
|
|
0.246261
(1.489121)
|
0.275402*
(1.657242)
|
0.326361**
(2.004394)
|
0.326372**
(2.000689)
|
BOR
|
|
|
0.736166*
(1.668727)
|
0.219180
(0.574093)
|
0.231719
(0.554164)
|
FTA
|
|
|
|
1.641467***
(4.475906)
|
1.665468***
(3.478030)
|
APEC
|
|
|
|
|
0.044916
(0.082993)
|
C
|
-8.607024**
(-2.448278)
|
-8.079963**
(-2.294449)
|
-12.97248***
(-2.853963)
|
1.745031
(0.348955)
|
1.651219
(0.318757)
|
修正的R2
|
0.555450
|
0.556544
|
0.558287
|
0.578381
|
0.576933
|
F统计量
|
206.7452
|
155.9942
|
125.8748
|
113.9455
|
97.23748
|
P值
|
0.000000
|
0.000000
|
0.000000
|
0.000000
|
0.000000
|
Note: the data in brackets are regression coefficients, and the data in brackets are t statistics. "*", "* *", "* *", "* *" were significant at 10%, 5% and 1% levels, respectively.
- The robustness test of regional bilateral trade model
There are many countries along the “Belt and Road”. In order to further illustrate the impact of economic scale, geographical distance and other factors on China's bilateral trade volume with various regions, the article will study the group of countries and conduct robustness tests on panel data in Central Asia, South Asia, Central & Eastern Europe, ASEAN and West Asia. The regression results are shown in Table 7.
Table 7 robust regression analysis of bilateral trade model
|
Central Asia
|
South Asia
|
Central & Eastern Europe
|
ASEAN
|
West Asia
|
LNGDPi
|
0.063991
(0.284467)
|
0.214647
(1.407646)
|
0.682336*
(14.77180)
|
0.845174*
(10.11135)
|
0.248225*
(3.776103)
|
LNGDPj
|
0.941409*
(3.549798)
|
0.903117*
(14.31794)
|
0.915398*
(25.45924)
|
0.660635*
(8.432675)
|
1.056731*
(9.861981)
|
LNDIS
|
-2.633330
(-0.658406)
|
-1.858127*
(-3.193092)
|
0.225623
(0.847972)
|
-1.357584***
(-1.744177)
|
-1.648303
(-0.799799)
|
LNWEF
|
0.167756
(0.331002)
|
1.627423*
(3.718248)
|
0.759197*
(5.782335)
|
0.903672**
(2.478173)
|
1.258311*
(3.928803)
|
C
|
19.62017
(0.591457)
|
8.439539***
(1.724144)
|
-13.65348*
(-5.714424)
|
-1.568486
(-0.248570)
|
6.519313
(0.358005)
|
ADjustedR2
|
0.298518
|
0.860002
|
0.870770
|
0.754835
|
0. 682457
|
Fstatistics
|
6.744970
|
100.8227
|
258.7343
|
84.89976
|
59.56509
|
PValue
|
0.000202
|
0.000000
|
0.000000
|
0.000000
|
0.000000
|
Note: the data in brackets are regression coefficients, and the data in brackets are t statistics. "*", "* *", "* *", "* *" were significant at 1%, 5% and 10% levels, respectively.
It can be seen from table 7 that except for Central Asia, the goodness of fit of bilateral trade models in other regions is relatively high, especially in South Asia and central and Eastern Europe, with the revised R2 values of 0.860 and 0.871 respectively. The significance level of P value in each region is more than 1%. In Central Asia and West Asia, the gap between China's GDP and the GDP of trading countries on the bilateral trade volume is very obvious. The regression coefficients of LNGDPi and LNGDPj in Central Asia are 0.064 and 0.941 respectively, while those in West Asia are 0.248 and 1.058 respectively. In the ASEAN region, China's GDP has a greater impact on the bilateral trade volume than the GDP of trading countries, with the regression coefficients of 0.845 and 0.661 respectively. Except for central and Eastern Europe, LNDIS has a negative impact on bilateral trade volume. The degree of hindrance of geographical distance to bilateral trade volume from small to large is as follows: ASEAN (- 1.358) < West Asia (- 1.648) < South Asia (- 1.858) < Central Asia (- 2.633). The quality of port infrastructure in each region has a positive effect on bilateral trade volume. For every 1% increase in LNWEF in South Asia, bilateral trade volume will increase by about 1.63%, which is significant at the level of 1%. For every 1% increase in LNWEF in Western Asia, LNTRA increases by 1.26%, which is also significant at the level of 1%. The regression coefficients of LNWEF in CEE and ASEAN are 0.759 and 0.904, which are significant at the level of 1% and 5% respectively.
4.2 Regression results of bilateral trade model
Both the overall regression results and the regional regression tests show that the bilateral trade model constructed in the article has a good fit, which can better reflect the economic scale, geographical distance, port infrastructure quality and three control variables (common border, FTA and APEC) impact on bilateral trade between China and countries along the “Belt and Road”. The specific analysis is as follows:
First, the scale of the economy has a positive effect on the growth of bilateral trade between China and countries along the “Belt and Road”. The article uses China's GDP and the GDP of trading countries to comprehensively measure the impact of economic scale on bilateral trade, and more accurately shows that the strength of a country's economic strength will have a significant stimulating effect on the country's trade. From the regression results, it can be found that the economic scale of the trading countries in Central Asia, South Asia, Central and Eastern Europe, and West Asia has a more obvious positive effect on the growth of bilateral trade. The main reason may be that these countries are importing countries for China’s exports. The huge demand capacity shows that the "Belt and Road" initiative meets the needs of the economic development of all countries and has a broad space for development.
Second, geographic distance has a negative effect on the growth of bilateral trade volume between China and the countries along the “Belt and Road”. Generally speaking, the farther the distance between the two countries is, the higher the various trade costs of trade exchanges between the two sides will be, and the less conducive to the development of bilateral trade. However, from the regression results of the article, the negative impact of geographic distance on trade is not absolute. The signs of geographic distance in models (1) ~ (3) are all opposite to expectations. The article believes that there are two main reasons: one is that with the advent of the Internet information age, people can freely conduct large-scale cross-border e-commerce transactions online, and the hindrance of geographical distance to bilateral trade is getting smaller and smaller or even negligible. Second, with the implementation of the “Belt and Road” project, China and its partner countries have become increasingly close in terms of facility connectivity. Not only have they signed a series of transport agreements involving sea, land, and air transport, but also successfully built railways and highways extending in all directions; making transportation more convenient and efficient, thereby reducing the adverse impact of geographic distance on bilateral trade.
Third, the quality of port infrastructure has a positive effect on the growth of bilateral trade between China and countries along the “Belt and Road”. Shipping has always been an important means of transportation of goods in the development of foreign trade by countries all over the world. With the rapid development of land transportation, air transportation and even multimodal transportation, the role of sea transportation in connecting countries' trade is becoming more and more important. The quality of port infrastructure is an important indicator that determines the efficiency of shipping. The regression results show that in Southeast Asia and West Asia, where shipping is developed, high-standard port infrastructure quality can bring good development benefits to the opening of regional trade.
Fourth, having common borders, signing FTAs and joining APEC have a positive effect on the growth of bilateral trade between China and countries along the “Belt and Road”. First of all, the countries with a common border have more in common in the cultural backgrounds, language, customs, etc., and it is easier to reach consensus in business exchanges, which creates a good cultural environment for the smooth development of bilateral trade. Secondly, signing an FTA can bring more preferential tariff policies to trading countries, eliminate the hindrance of tariff and non-tariff barriers as much as possible, and create a stable trading environment for the growth of bilateral trade. Finally, joining APEC shows that member states share common economic development objectives and social development aspirations, and create a harmonious institutional environment for the promotion of bilateral trade.