5.1 Unit root test
Before using Granger causality test, we have to process the unit-root test to examine whether the data is stationary. Table 4 to Table 6 shows the descriptive statistics.
We test ALLC, LLC, SLC and GDP variable by ADF test. The test has to be considered the lag length, we choice the lag length by using AIC (Akaike info Criterion). The null hypothesis was set as H0: the variable has a unit root. The results can be seen in Table 7 and Table 8.
In Table 7 the second column represent ADF test of ALLC, at 5% significant level, the liquidity creation growth of all bank (ALLC) of six countries (Canada, Italy, South Korea, Netherlands, New Zealand and USA) can reject the null hypothesis, it means that the unit root does not exist.
To ensure that all series can be stationary, we construct a process of first-difference and retest the ADF test. Table 8 second column shows that all the new series except Australia and Spain can reject the null hypothesis4. But Australia and Spain are stationary in second-difference.
Repeat the process; we test LLC by level and first-differencing. Table 7 third column indicates that Italy, South Korea, Netherlands and USA should be retest by
4 The ALLC of Australia and Spain is stationary in second-difference. Each t-statistic is -3.45897 and -6.5693.The p-value is 0.0378 of Australia, and 0.0002 of Spain. It‘s significant reject the null hypothesis.
first difference. These four countries are stationary in first-differencing as Table 8 third column.
We also test SLC whether the unit root exist or not. Table 7 and Table 8 fourth column indicate that all the variables are stationary in first-differencing except Italy.
SLC of Italy is stationary in second-difference. The p-value is 0.047. It‘s significant reject the null hypothesis
Then we examine the GDP of each country. In Table 7, only Germany and USA can be rejected at 1% significant level. For this reason, we have to use first-differencing and test again. We can find that the GDP growth of 10 countries can be stationary except Australia in Table 8. GDP of Italy is stationary in second-difference. The t-statistic and p-value are -3.11279 and 0.0616. It‘s significant reject the null hypothesis.
We also use Phillips-Perron test to examine the stationary. And we get a consistence result with ADF unit root test. Table 9 illustrates some variables are stationary at level. And Table 10 represents the result of first-differencing.
5.2 Co-integration test
In this section, we test the long-run relationship for four parts in these 10 countries. According the result of ADF test, we find there are not all the variable being stationary in levels.
First we test the co-integration relationship between ALLC and GDP. In this part, only France should be test whether this relationship exists. We use the trace test and maximum eigenvalue test the co-integration, the conclusions are the same and the test
power is less than trace test, so we only explain the results by trace test.
Table 11 illustrate that the trace test rejects the H0. It means that there is one co-integration relationship at least exist in long-term
Then we test the co-integration of LLC and GDP in Canada, France and New Zealand. For Canada and France, we get a result which is there is a co-integration relationship at least in the countries.
But New Zealand doesn‘t have the co-integration relationship. The results show in Table 12. So when we test the Granger causality, Canada and France should use VEC model, and New Zealand is used the VAR model.
Third, the co-integration between SLC and GDP are showing as Table 13. When we test the co-integration test between SLC and GDP, only test France. And the result indicates there is still a co-integration relationship at least, so we use VEC model for France.
Finally, we test the relationship between LLC and SLC. Australia and France should do this test. But result shows there are not co-integration relationship; we only use the VAR model for the 10 countries.
5.3 Granger Causality test
The process of Granger causality test can help us find the relationship between two variables. There are three possible outcomes; one is prior indicator, both variables have a feedback relationship or they are dependent.
However, Granger causality test only identifies the lead and lag relationship. The relationship just can be a reference for forecasting research. There are different results
in different countries.
We use the stationary series data to construct a VAR model. And according to the model, we choose the optimal lag length. Then we test the Granger causality. The hypothesis is written as:
H0: GDP does not Granger Cause ALLC H1: ALLC does not Granger Cause GDP
If the F- statistic over test statistic, we can reject the hypothesis. It means if H0 is rejected, GDP would cause ALLC. If H1 is rejected, then ALLC would cause GDP.
Not only these two result but there is a feedback effect, which is represent two variables will cause the other, on the other hand, H0 and H1 exist at the same time.
Similarly, we can change the H0 and H1 test the causality of LLC and GDP, and SLC and GDP.
5.3.1 ALLC and GDP
In this test, we find that there are different results in different countries. It shows in the Table 14
First, ALLC and GDP are independent in Australia, Canada, Germany, Italy, and Netherlands. The F-statistic and corresponding p-value is significant non-reject H0
and H1. This means more liquidity creation could not cause the economic growth. In a bloom economic environment, it doesn‘t matter the banker‘s willing of lending money to firm.
Second, GDP can cause ALLC in Spain. The p-value (0.0919 is significant lower
than 10% significance level.
Third, France, South Korea, Zealand and USA, the H0 and H1 can be rejected at the same time, ALLC and GDP leads each other. It means there are some variable we don‘t observe cause the two variables.
For France, we use VEC model. ALLC will cause GDP in France at long-term relationship. The t-statistic (11.9932) is very high. In short-term, not only ALLC with lag 1 can cause GDP, but GDP with lag 1 cause ALLC. It has a significant feedback effect in France.
From Bank-based and Market-based perspective, we can categorize these 10 countries by sort. The countries of Bank-based are France, Germany, Italy, New Zealand and Spain, and the market-based countries are Australia, Canada, South Korea, Netherlands, and USA. We find that there is not a significance difference between these two systems.
5.3.2 LLC and GDP
According the result of co-integration test, we find all countries should use VAR model except Canada and France. The result in Table 16 still matches the all outcome when we test ALLC.
The variables are independence in Australia, Italy, Netherlands and Spain. The p-value cannot reject Ho at 10%, 5% and 1% significance level.
It has a feedback effect in South Korea, New Zealand and USA. And the result shows us that change of large liquidity creation will lead the economic growth in
Germany. Table 17 Panel A is the model of each country.
We test Canada and France by VEC model. Seen Table 16, in Canada, the result is consistent with ALLC. There is not any Granger causality relationship in the country.
LLC also cause GDP in France at long-term relationship. However, there is a feedback effect in short-run relationship.
About Granger causality of LLC and economic growth, we find a consistence result with ALLC, it may because most of liquidity creation are created by these large bank. We classify the ratio of each country‘s large banks liquidity creation to their all banks liquidity creation in Table 18.
We observed a phenomenon that the ratios are almost higher than 80% except New Zealand. It explains why these two tests can get almost consistent result.
5.3.3 SLC and GDP
We use VAR model to test for these 10 countries except the France which has the co-integration relationship.
Table 19 shows that there are not lead/lag relationship in all countries. The two variables are independent in all the countries. The reason is that the sampling banks are the smallest at the bottom in the bank industry. They are hard to cause some important influence for the industry. Table 20 shows the model.
5.3.4 LLC Granger Cause SLC?
In most industries, because large firms‘ resource and information are more sufficient than small firms, they often lead the smaller firms. In bank industry, the bank be an intermediation role, they have to get the investor and lender‘s needs. On the other word, they can get more market information to set up a liquidity policy, and then the smaller banks just are follower.
In this part, we test the relationship between large bank and small bank. Table 21 is the result of Granger causality test and we classify the VAR model in Table 22.
Surprisingly, we find that there is not significant evidence can tell us that large bank can be a leader in America; the p-value is less than 10%. But Small bank even can be a leader in France, seeing Table 21.