4.1.2. Independent Variable Comparison

Comparing independent variables of different types of banks can give us their specific characteristics. This paper finds that policy banks have less liquidity and profitability; correspondingly, commercial banks have a greater advantage in these two areas. Studying commercial banks further shows that large-scale banks present less liquidity, lower credit risks, and more adequate regulatory capital for whole periods.

(1) Policy banks versus commercial banks

The most significant differences between policy banks and commercial banks are in liquidity and performance, especially liquidity (see Tables 4 and 5). The average liquidity ratio of policy banks can be around five times higher than that of commercial banks (364.4% versus 77.1%). The gap of the liquidity ratio became even wider after 2017, and the average performance between policy banks and commercial banks changed significantly. Variable risk exposure and regulatory capital were similar.


**Table 4.** Bank-specific determinant variable comparison (pre-2017Q4).

Data were collected from Bloomberg, Wind, the banks' financial reports, and the China Securitization Analytics website.


**Table 5.** Bank-specific determinant variable comparison (post-2017Q4).

Data were collected from Bloomberg, Wind, the banks' financial reports, and the China Securitization Analytics website.

In order to investigate liquidity further (see Tables 6 and 7), the liquidity variables were divided by (1) net loans/deposits and short-term funding (the ND ratio) and (2) liquid assets/deposits and short-term funding (the LD ratio). The ND ratio of policy banks is much higher than that of commercial banks (315.6% versus 51.7%), which indicates paradoxically that the loans provided by policy banks are around three times greater than their own deposits and short-term funding, which could result in poor liquidity. Even though policy banks on average acquire more liquidity assets compared to commercial banks (48.8% versus 25.2%), they still struggle with poor liquidity because of the massive amount of loans. After 2017Q4, the liquidity issue of policy banks was more serious. The ND ratio of policy banks was about 12 times higher than that of commercial banks.


**Table 6.** Bank-specific determinant variable comparison (pre-2017Q4).

Data were collected from Bloomberg, Wind, the banks' financial reports, and China Securitization Analytics website.

Both (7) the cost-to-income ratio and (8) the return on assets (ROA) ratio were used to measure bank performance. The difference in the performance ratios between the policy banks and commercial banks is mainly caused by the ROA rather than the cost-to-income ratio. The cost-to-income ratio of the two types of bank are similar (44.1% in policy banks versus 41.1% in commercial banks). However, the cost-to-income ratio of policy banks became much higher than that of commercial banks after 2017Q4, which means that policy banks have higher operating costs. However, the mean of the ROA of commercial banks is much higher than that of policy banks. The ROA of commercial banks is 1.012, which is about 25% higher than that of policy banks. The high ROA of commercial banks reflects that commercial banks have a greater advantage in profitability than policy banks. This also indicates the different operating visions of these two types of bank; policy banks are for policy-related lending, while commercial banks pursue higher profitability.


**Table 7.** Bank-specific determinant variable comparison (post-2017Q4).

Data were collected from Bloomberg, Wind, the banks' financial reports, and the China Securitization Analytics website.

#### (2) (City/rural commercial banks versus national commercial banks

Commercial banks are a large part of our sample, which were divided into two types and analyzed further. On average, commercial banks with larger-scale assets presented less liquidity, lower credit risks, more adequate regulatory capital, and better performance before 2017Q4 (see Tables 6 and 7). However, the city/rural commercial banks had an advantage in credit risks over national commercial banks after 2017, which improved their performance.

The ratios employed to measure the bank's credit risks are (3) loan loss reserves/total loans (the LL ratio) and (4) impaired loans/total loans (the IT ratio). National commercial banks had a greater advantage in credit risk management compared with city/rural commercial banks before 2017. The LL ratio and the IT ratio of the city/rural commercial banks were much higher than those of the national commercial bank, which was as high as 3.6%. This indicates that banks with a larger scale are better at risk management. However, the LL ratio of the national commercial banks increased significantly and became much higher than that of the city/rural commercial banks, which caused those banks to lose their advantage in risk management.

Both the CIR and ROA variables, as banking efficiency or performance measures, show that the city/rural commercial banks' performance was worse (43.4% and 101%) than that of the national commercial banks during the first period. Hence, banks with large-scale assets tend to have better performance. However, the profitability of both kinds of bank changed after 2017Q4, and their profitability tended to be similar.

#### 4.1.3. Univariate Analysis

The previous analysis is based on part of an independent variable comparison. This section analyzes how those characteristics affect their securitization (dependent variables).

#### (1) Policy banks versus commercial banks

The securitization transaction volume of commercial banks is much higher than that of policy banks for the two periods. The mean percentages of the transaction volume to total assets are 0.35% and 0.28% for the commercial banks, as opposed to 0.16% and 0.14% for the policy banks (see Tables 6 and 7). The previous section shows that liquidity and performance are the two major different variables between policy banks and commercial banks for the whole period. Therefore, the liquidity and performance of banks might be two significant determinants that affect loan securitization. Entities resorting to securitization are net borrowers of funds in the interbank market and are seeking to improve its financial position.

Comparing (1) the ND ratio, (2) the LD ratio (liquidity measures), (7) the CIR, and (8) the ROA (performance measures), with higher liquidity and performance, banks acquire

securitization issuances. The other two determinants, risk exposure and regulatory capital, also reflect the relationship with bank loan securitization. Risk exposure and regulatory capital are positively related to securitization transaction volume, even though their effects are limited.

(2) City/rural commercial banks versus national commercial banks

The transaction volume to the total assets in commercial banks increased with their decreasing asset scale for the whole period, even though the amount of securitization issuance rose with a larger asset scale. The percentages of the transaction volume to the total assets regarding city/rural commercial banks was the largest (0.78% and 0.64%), much larger than those of national commercial banks (0.21% and 0.15%).

The previous section indicates that liquidity and performance are also significantly different variables for the two types of bank. Thus, this paper considers the difference in the securitization transaction volume to the total assets because of the important liquidity and performance before 2017Q4. After that, the liquidity and regulatory capital were the two significantly different variables, so the motivations for the securitization of commercial banks changed. Improving liquidity and regulatory capital arbitrage is expected to be the motivation of securitization after 2017Q4. The subsequent analysis will confirm whether this variable is statistically significant in the model.

#### *4.2. Multivariate Analysis*

#### 4.2.1. Groups of Bank Samples

This paper focuses first on regression on all bank levels, followed by research on types of bank. The sample of banks is divided into three main groups, namely, (1) whole banks, (2) commercial banks, and (3) national commercial banks (see Table 8). Whole banks consist of all banks (policy banks and commercial banks); commercial banks are composed of city/rural commercial bank and national commercial banks. The national commercial banks are the last group studied.

#### **Table 8.** Types of bank group.


4.2.2. Results of Four-Variable Regression

This paper examines four variables using a fixed effects model and a random effects model. According to the Hausman test, the probability of all results are lower than 95%, which means that the composite error term is correlated with all of the explanatory variables. Thus, a fixed model is more appropriate. The following analysis is based on the results of the fixed effects model (see Tables 9–12).

**Table 9.** Regression results of four variables in *t* − 1 (pre-2017Q4).



**Table 9.** *Cont.*

Data were collected from Bloomberg, Wind, the banks' financial reports, and the China Securitization Analytics website; *t*-statistics are in parentheses. \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.10.



Data were collected from Bloomberg, Wind, the banks' financial reports, and the China Securitization Analytics website; *t*-statistics are in parentheses. \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.10.

**Table 11.** Regression results of eight variables in *t* − 1 (pre-2017).


Data were collected from Bloomberg, Wind, the banks' financial reports, and the China Securitization Analytics website; *t*-statistics are in parentheses. \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.10.


**Table 12.** Regression results of eight variables in *t* − 1 (post-2017).

Data were collected from Bloomberg, Wind, the banks' financial reports, and the China Securitization Analytics website; *t*-statistics are in parentheses. \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.10.

#### (A) All banks

Liquidity and risk exposure were the two important determinants of securitization in China's banking sector before 2017Q4, showing a confidence level of more than 99%. Compared with liquidity, risk exposure presents more a significant effect on loan securitization transaction volume, because the probability of securitizing increases given more variation in the dependent variable. When a bank's risk exposure increased by one unit, the probability that a bank will opt for securities increased by 17.4% when the other variables were held constant. Risk exposure had a positive effect on securitization. The liquidity effect on loan securitization was limited. A one-unit liquidity change only resulted in a 0.1% securitization transaction volume adjustment. The higher risk exposure motivated banks to issue more loan securities. Additionally, banks with a lower liquidity could raise liquidity and funding via securitization.

Risk exposure was an important determinant for securitization only after 2017Q4, showing a confidence level of more than 90%. When a bank's risk exposure increased by one unit, the probability that a bank will opt for securities increased only by 2.4% when the other variables were held constant, which is much lower than that before 2017Q4. Additionally, improving liquidity was not a determinant of securitization for all banks.

#### (B) Commercial banks

Consistent with all banks, liquidity and risk exposure were still the two important determinants of the loan securitization transaction volume. The risk exposure ratio was significant at a 99% confidence interval and with an obvious effect on the securitization transaction volume (a one-unit risk exposure rise corresponds to a 15.7% change in securitization transaction volume). Liquidity was only in the 90% confidence interval, so it is not as important as risk exposure.

The motivation for commercial banks' securitization issuance was similar to the other banks. Risk exposure was the only determinant after 2017Q4.

#### (C) National commercial banks

The determinants of securitization in national commercial banks were completely different from the previous two groups before 2017Q4. Capital requirements and profitability were two important determinants in this group. Capital requirements, compared with performance, was more significant with respect to securitization. When a bank's regularity capital decreased by one unit, the probability that a bank would opt to securitize

increased by 2.7%. With a lower regularity capital, the banks acquired a higher securitization transaction volume, which could reflect securitization as a way to search for new profit opportunities. The profitability variable was statistically significant, indicating that banks are using securitization to raise their performance, but its effects on national commercial banks are limited (only a 0.1% regression coefficient).

However, capital requirements and profitability were not the determinants of securitization after 2017Q4. The table shows that risk exposure was still the determinant for national commercial banks, which was significant at a 99% confidence interval.

#### 4.2.3. Results of Eight-Variable Regression

In Tables 11 and 12, as with the four-variable regression analysis, both the random effects approach and the fixed effects approach were applied in this regression. According to the Hausman test probability, the fixed effects model is valid.

To further confirm the findings, eight-variable multivariate analysis was conducted. Each determinant was evaluated by two proxies, introduced in the methodology section. If both of two variables were in a confidence interval greater than 90%, the determinant was considered to drive securitization issuances. Additionally, if only one variable was statistically significant in relation to the transaction volume, its influence on securitization was concluded.

#### (A) All banks

Liquidity and risk exposure were the main drivers of loan securitization in the Chinese banking sector, which is basically consistent with previous results, but performance was also a significant driver of securitization in the eight-variable regression before 2017Q4. Specifically, (3) the LT ratio and (4) the IT ratio as risk exposure measures were statistically significant. The coefficients of (3) the LT ratio and (4) the IT ratio were 10.9% and 36.6%, respectively, and appear to exert the most influence on the probability that a bank opts to securitize, compared to the other variables. (1) The ND ratio as a proxy of liquidity indicates statistical significance at the 5% level. While the liquidity effect was limited, the one-unit ND ratio rise only improved securitization truncation volume by 0.1% in all banks. Even though this regression result indicates that securitization can be used as a way to improve a bank's performance, the coefficient of this ratio is too small, so its effect is limited. (8) The ROA ratio variable as a profitability measure is the least statistically significant determinant among the four basic determinants proposed in the literature.

After 2017Q4, reducing capital requirement and risk exposure was the main determinate for the whole banks. Specifically, (4) the LT ratio was statistically significant, but the coefficients of that ratio were much lower than that before 2017Q4. Both (5) the TETA ratio and (6) the ROA ratio are related to loan securitization issuance. However, they are both significant at a 90% confidence interval.

#### (B) Commercial banks

In the group of commercial banks, all four determinants affected loan securitization before 2017Q4 but to varying extents. Risk exposure, compared with the other determinants, was the most significant for securitization. Two variables, (3) the LL ratio and (4) the LT ratio, presented statistical significance at the 99% confidence level. The coefficients of both ratios were also the highest compared to the other variables—11.1% and 30.2%, respectively. Liquidity was the second most significant determinant. (1) The LD ratio measuring liquidity was related to loan securitization. These two determinants are consistent with the fourvariable regression. The capital requirement and performance were statistically significant, which also motivates banks to securitize part of its portfolio, but not as significant as risk exposure and liquidity. (6) The tier one ratio (capital requirement measures) and (8) the ROA ratio (profitability measures) were statistically significant, but only in the 90% confidence interval, so they were the least statistically significant. This might explain why neither of them were significant in relation to loan securitization in the four-variable regression. The coefficient value of (8) ROA (performance proxy) was close to zero. Using

securitization as a mechanism for improving a commercial bank's performance does not seem to be very efficient.

The motivation for the securitization issuance of commercial banks was only regulatory capital arbitrage after 2017Q4. Reducing risk exposure and increasing liquidity and performance were no longer determinants of securitization issuances. (6) The tier one ratio (capital requirement measures) was statistically significant but only in the 95% confidence interval.

(C) National commercial banks

Regularity capital was the only driver of securitization activities in national commercial banks. (5) The TETA ratio measuring capital requirement was the only variable with statistical significance. Profitability was an important determinant for securitization in the four-variable regression, but (7) the CIR and (8) ROA variables measured as bank profitability did not reach statistical significance in the eight-variable regression. This leads to a new conclusion: Regulatory capital, rather than performance, is the only determinant that appears to exert the most influence on loan securitization. National commercial banks could lower their regularity capital (regulatory capital arbitrage) via securitization. Interestingly, risk exposure was no longer a significant determinant for a bank's securitization decisions in the national commercial banks. This is completely different from all other banks.

Risk exposure was the other determinant of securitization issuance after 2017Q4. The risk exposure and regularity capital were two main drivers of securitization activities in national commercial banks. Specifically, (4) the IT ratio, measured as a bank's risk exposure, reached statistical significance at the 99% confidence level. (7) The CIR and (8) ROA variables were also statistically significant.

In summary, the results of the eight-variable regression are basically consistent with the four-variable regression, but they also revealed some new important determinants for securitization. Specifically, securitization transaction was motivated by both risk exposure and liquidity, risk exposure especially in the first period, but was still motivated by risk exposure after 2017Q4. However, the eight-variable regression shows that performance was another significant determinant for securitization, even though its effects were limited before 2017Q4.

#### *4.3. Results of Varying Types of Banks*

#### 4.3.1. Derivations from Regression Results

The above findings only reflect how these determinants affect securitization decisions in varying bank groups, but it is difficult to indicate how determinants influence varying types of banks, not including national commercial banks. This can be safely deduced by comparing different bank groups (see Table 13). Specifically, city/rural commercial banks can be deduced through a comparison of the *p*-value and coefficients of commercial banks and national commercial banks. For example, if the former regression probability of a hypothesis variable (*p*-value) (commercial banks) is higher than that of the national commercial banks, national commercial banks can be considered to have contributed to an increased *p*-value. If the variable regression probability is the same or similar, their coefficients will be compared and their influence inferred. Policy banks are also analyzed according to this methodology.


**Table 13.** Types of bank.


National commercial banks and their change were analyzed in the four-variable regression section, so we do not need to compare and discuss their important determinants. Liquidity was the only determinant before 2017Q4, but risk exposure became the main motivation for securitization issuance after 2017Q4.

(2) City/rural commercial banks

In city/rural commercial banks, liquidity and risk exposure were the main determinants for securitization before 2017Q4. As per the previous analysis, capital requirement was the only significant determinant in national commercial banks. In other words, liquidity and risk exposure were not related to national commercial banks' securitization. However, these two variables were statistically significant in all commercial banks. This implies that city/rural commercial bank liquidity and risk exposure are related to the dependent variable and result in the statistical significance of all commercial banks.

However, liquidity and risk exposure were not the drivers of loan securitization issuance after 2017Q4. The *p*-value of national commercial banks was in the 99% confidence interval, which was higher than that of commercial banks (in the 95% confidence interval). This implies that city/rural commercial banks have no motivations for securitization issuance.

(3) Policy banks

Liquidity and risk exposure were significant determinants motivating policy banks' securitization before 2017Q4. The risk exposure *p*-value in the all-bank regression was in the 99% confidence interval, and this was found for the national commercial banks as well. Thus, their coefficients were further compared. The coefficient of risk exposure variables in all banks was higher than that of all commercial banks (17.4% versus 15.7%). The risk exposure of the policy banks could influence their securitization decisions and raise the corresponding coefficient in the all-bank regression. The regression probability of liquidity in all banks is higher than that in all commercial banks. Thus, the liquidity of policy banks was also a significant determinant for their securitization transaction and improved the probability in the all-bank regression.

However, it is hard to infer specific drivers by comparing *p*-values of the commercial bank group and the all-bank group. Due to the lower *p*-value of the all-bank group, we inferred that there is no motivation for securitization issuance. This outcome is the same in the case of rural/city commercial banks.


More detail about national commercial bank securitization determinants can be found in Section 4.2.3.

(2) City/rural commercial banks

Liquidity, risk exposure, and profitability are three important determinants for securitization transaction volume in city/rural commercial banks. This is because these three determinants are not statistically significant in the former group but present contrary outcomes in the commercial bank group. The statistical significance comes from the effect of city/rural commercial banks. The (5) TETA ratio is also statistically significant in the sample of commercial banks. However, its *p*-value is lower than that of the national commercial bank group, confirming that capital requirement is a significant determinant in city/rural commercial banks.

Only the regulatory capital arbitrage is inferred to have been an important determinant after 2017Q4. The (5) TETA ratio was also statistically significant in the sample of commercial banks and is the same as that of the national commercial banks in the 99%

confidence interval. The coefficient was higher than that of the commercial bank group, which could imply that regulatory capital arbitrage was the main motivation for city/rural commercial banks to issue securities.

(3) Policy banks

Risk exposure was inferred to have been an important determinant in policy banks before 2017Q4. The (1) ND ratio (liquidity measures), the two variables (3) and (4) of risk exposure, and (8) the ROA ratio (profitability measures) were statistically significant in all banks, but the regression coefficients of (1) and (8) were smaller or equal to the former groups, which makes it difficult to prove that liquidity and profitability were two important determinants of policy bank securitization issuance. The *p*-values of variables (3) and (4) of all banks were the same as those of the commercial bank group. Although the coefficient of (3) was lower than that of the sample of commercial banks (10.9% versus 11.1%), the coefficient of (4) in the sample was much higher than that of the commercial bank sample (36.6% versus 30.2%). Therefore, the effect of risk exposure in all banks was greater than that of the commercial bank group. The risk exposure affected policy bank securitization and improved the corresponding coefficient.

Risk exposure and regulatory capital arbitrage were inferred to be two main determinants in policy banks after 2017Q4. Regarding risk exposure, (3) the LT ratio was statistically significant in the 90% confidence interval in the all-bank group, but there was no statistical significance in the commercial bank group. We conclude that national policy banks contributed to an increased *p*-value. In the same way, it can also be inferred, by comparing (5) TETA ratios, that the regulatory capital arbitrage was the other main determinant.
