Literature

Our work is mainly related to two strands of literature. First is the risk-taking channel of monetary policy transmission. Borio and Zhu [17], among others, articulate the connection between the interest rate level and bank risk-taking activity. This theory claims that the policy interest rate is not only the cost of refinancing for commercial banks, but it is also a signal of the market risk level. Commercial banks infer a lower market risk level from monetary easing and thus are inclined to increase leverage ratio and invest in a riskier asset portfolio. Much empirical literature [19–22] supports this theory. Some studies [23–25] investigate the risk-taking channel of monetary policy in emerging markets, but few consider this effect in the interbank market in China. Some papers discuss about bank risk-taking behavior in China, but they mainly study the connection between bank risk-taking behavior and fintech, bank efficiency, law enforcement and bank capital structure [26–28]. Moreover, even fewer consider the risk-taking channel of SLF in the interbank market. We study the interbank effect of SLF and empirically show that the risk-taking channel not only plays a role but is the dominant mechanism in the process.

In addition to the risk-taking behavior in general, we examine how and to what extent the financial heterogeneity of banks affects their risk-taking behavior. Thus another strand of related literature studies the heterogeneous reaction of commercial banks to monetary policy. Some works [29, 30] claim that monetary policy transmission is influenced by capital regulation on banks. Kashyap and Stein [31] and Bernanke and Lown [32], among others, show that when the capital requirement is binding for commercial banks and it is costly for them to issue non-reservable liabilities or equity, expansive monetary policy may not stimulate lending. This theory is called Capital Regulation Theory, and it implies less reaction of banks with worse accounting indices to monetary policy. Some other works [33] claim that it is more costly for banks with worse accounting indices to replenish their reserve, so these banks depend more on the liquidity supply from the central bank and tend to react greater to monetary policy. This is the Market Friction Theory. Finally, Van den Heuvel [34] emphasizes the Interest Income Theory, which is the impact that monetary policy has on traditional bank revenue, i.e., interest income. Monetary expansion tends to steepen the yield curve when the interest rate level is not close to the zero lower bound, and it increases the traditional revenue of banks (Brunnermeier and Koby [unpublished]). Hence, the interest income theory predicts marginally better liquidity conditions of banks that rely more on interest income.

Hypotheses

According to the risk-taking channel theory, the SLF affects both the demand and supply of reserves in the interbank market. The SLF policy stance not only implies the cost and accessibility of refinancing (liquidity supply), but it also serves as a signal released by the central bank, which affects risk preference and thereby the asset portfolio allocation of banks, and finally the liquidity demand. Further, bank accounting heterogeneity is a significant factor when we consider the reserve and capital acquisition, as well as the traditional revenue of banks. These can lead to differential reactions of banks to SLF. As summarized in Fig 2, we study both the supply and demand effect of SLF on the interbank market. For the demand side (the lower part of the graph), a higher outstanding SLF quantity encourages risk-taking behavior of banks, so their willingness corporate lending, while interbank lending decreases, which cumulatively leads to higher demand for interbank liquidity. For the supply side (the upper part of the graph), our focus is on how bank financial heterogeneity can affect SLF transmission.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. SLF transmission mechanism.

https://doi.org/10.1371/journal.pone.0284470.g002

To study the expectation effect of SLF, first, we argue the following:

1. Hypothesis 1 An outstanding SLF quantity expansion encourages bank risk-taking behavior, while its contraction restrains bank risk-taking behavior.

Hypothesis (1) is a natural inference from the traditional risk-taking theory in the context of SLF. Unfortunately, it only refers to the increase in liquidity demand induced by SLF quantity expansion; it does not necessarily imply greater liquidity tension. To complete the logical chain, we need to illustrate that a riskier portfolio of banks tends to raise the interbank rate up. We hereby propose the following:

1. Hypothesis 2 A higher level of bank risk-taking activity aggravates interbank lending market liquidity tension, while a lower level of risk-taking activity alleviates the liquidity tension.

Banks with heterogeneous characteristics face diverse liquidity situations and reveal different risk preferences. Financial indices and ownership directly affect the funding source diversity and financing costs of banks, both of which affect the substitutability of SLF. Thus, we propose the following:

1. Hypothesis 3 The financial heterogeneity of banks makes them perform differentially in the SLF risk-taking encouragement process.

Econometric strategies

Although some works verify the risk-taking encouragement effect of other monetary policy tools, to verify hypothesis (1), we conduct the following regression: (1) where z_it is the z-score of bank i at period t, the proxy of bank risk measurement, ω_i is the bank individual fixed-effect term, and θ_t is the annual time fixed effect. slf_t−1 is the lagged outstanding SLF quantity, and MP_mt−1 denotes the lagged measurement of the other two monetary policy tools, OMO (m = 1) and Reserve (m = 2). X_nit represents accounting indices of banks, including tier 1 ratio (n = 1), net interest rate income ratio (n = 2), non-performing loans ratio (n = 3), operating income growth rate (n = 4), and asset growth rate (n = 5). The macro control variable Γ_t is the real GDP growth rate. ε_it is the error term. The significance of δ determines whether SLF has significant impact on bank risk-taking behavior, and the sign of δ determines the direction of the impact.

To verify hypothesis (2), we adopt the following regression equation: (2) where r_it−1 is the lagged term of the dependent variable r_it (the interbank lending market liquidity tension measured by the spread between the 1M Shanghai interbank offered rate, or Shibor, and interest rate on excess reserves, or IOER). The significance of parameter ζ determines whether risk-taking activity has an impact on the interbank bid rate, and the sign of ζ reveals the direction of the impact.

When we only examine the liquidity demand and supply of commercial banks, the greater risk appetite of banks increases the liquidity demand and reduces liquidity supply in the interbank market. Hence, combining hypothesis (1) and (2) explains the observed anomaly.

Hypothesis (3) indicates that bank heterogeneity influences its reaction to SLF risk-taking encouragement, while hypothesis (2) shows that the risk-taking behavior influences a bank’s interbank bid rate. Consequently, we can infer that bank financial heterogeneity affects a bank’s bid quoting behavior in the interbank lending market. This inference can be examined further with regression: (3) where . We include the sum of the interaction terms of accounting indices and all of the monetary policy tools, as well as the sum of triple interaction terms of MP ⋅ niir ⋅ cap. The changes of bank capital position impact the bank interest income and therefore influence the bid quotations. The triple interaction term controls for this indirect effect. According to the capital regulation theory (CRT), market friction theory (MFT), and interest income theory (IIT), which are mentioned in the literature, the expected signs for SLF interaction term parameters are listed in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Bank heterogenous reaction to SLF.

https://doi.org/10.1371/journal.pone.0284470.t001

As for the estimation strategy, normally, several-term-lagged variables are employed as instrument variables to estimate dynamic panel models with the GMM. Nevertheless, a long panel data set is used in this paper, and according to Arellano and Bover [35], IV is not necessary. Moreover, to avoid too much loss of the degree of freedom, we neglect the LSDV estimators, instead, Eqs (3) and (4) are estimated with FE estimators.

Endogeneity

Regarding econometric strategies, we should consider the possible endogeneity problem in our regressions. In all of the above regression equations, monetary policy variables are in a one-period lagged form, so the possible two-way causality between the interbank market rate/bank risk-taking behavior and monetary policy does not concern us. In addition, the interest rates on repos with rate securities as pledges for deposit-taking institutions (the DR rates), instead of Shibor, are considered as the potential market benchmark interest rates by the People’s Bank of China (China Monetary Policy Report, Quarter Three, 2016). Thus, Shibor does not directly affect the SLF policy decisions. Therefore, we do not anticipate the endogeneity problem in our regressions.

Data sources

We employ data from multiple sources. For time-series data, Shibor is from the official data set of the Shanghai Interbank Offered Rate (The URL address of Shibor home website is <http://www.shibor.org/shibor/web/html/index.html>.). while monetary policy stance variables and real GDP growth rate are from the WIND database. For panel data, the Shibor bid quotes of banks are from the same data set as the Shibor time series above, while the bank accounting indices are from the WIND database and originally from the consolidated financial statements of all of the listed quotation banks. Moreover, 15 out of 18 quotation banks are included in our sample set because HSBC (China), China Guangdong Bank, and China Development Bank are not listed either in the Shanghai or Shenzhen Stock Exchange, and they are thus excluded. For China Minsheng Bank, we use the bid rate from Q3 2017 to Q1 2022 because it was not a quotation bank until Q3 2017. For the full list of quotation banks, see S1 Table in the Supporting information. All of the data are quarterly. The duration is from Q1 2013 to Q1 2022, 37 quarters in total (19 quarters for China Minsheng Bank). For basic statistics of macro variables, see S2 Table.

In the sample set, banks are divided into two categories, state-owned commercial banks (SOCBs) and non-state-owned commercial banks (NSOCBs). The Industrial and Commercial Bank of China, Agricultural Bank of China, Bank of China, China Construction Bank, and Bank of Communication are SOCBs, while the others are NSOCBs. For the statistics of NSOCBs see S3 Table; for those of SOCBs, see S4 Table.

Now, we focus on the time-series graph of the liquidity tension of the interbank lending market measured by 1-month-Shibor-IOER spread. (The interbank-risk-free-rate spread is a common measurement of interbank liquidity conditions, such as Christensen et al. [7], Angelini et al. [36], and Albertazzi et al. [37]). See S1 Fig. For the graph of outstanding quantity of SLF during the same period, see S2 Fig. For the bid rate curves of all of the banks, see S3 Fig, they share a similar pattern.

Variables

Three types of variables are included in our models.

• Dependent variables. For the SVAR model, we use liq_t, the quarterly-averaged 1-month Shibor net interest rate on excess reserve (IOER), to measure the interbank market liquidity tension. For the panel data model, we use r_it, the spread between quarterly-averaged 1-month Shibor bids and IOER. To verify the different risk-taking behavior of SOCB and NSOCB, we employ z-score as the proxy of risk preference. The z-score, which is used in many studies [38, 39] as a bank risk index, is defined as z ≡ σ_ROA/(ROA + CA), where σ_ROA is the standard deviation of ROA, and CA is the capital adequacy ratio. (Note that the capital adequacy ratio (CA) is different from the tier 1 ratio). Although the z-score actually measures the de facto risk level of a bank, it reveals the risk preference of the bank.
• Explanatory variables. (a) Accounting index. Our models include the tier 1 ratio (Cap), net interest income ratio (NIIR), non-performing loans ratio (NPL), growth rate of operating income (OI), and asset growth rate (Asset). (b) Monetary policy stance. Besides our analysis object SLF, two more monetary policy tools are included to serve as control variables: the open market operation (OMO) representing traditional tools and the required deposit reserve ratio of financial institutions (Reserve) representing structural tools. SLF is captured with the outstanding quantity. OMO is measured by the net open market liquidity supply.
• Macro control variable. For the panel data model, we use the GDP growth rate (GDP) to control for the market liquidity condition changes.

All of the interest rates, growth rates, and ratios are in percentage, and the units of the outstanding SLF and OMO quantity are 10 trillion CNY.

The anomaly

Before the panel data regression, to preliminarily analyze the SLF effect on the interbank rate, a structural vector autoregression (SVAR) model is employed. All of the macro variables in the regression are stationary time series according to the ADF unit root test (see S5 Table). A VAR system of the SLF and the interbank liquidity tension (liq_t) is first established, the optimal lagged rank (see S6 Table) of the VAR is 1, and the VAR system is stationary (for the unit circle graph, see S4 Fig, for the eigenvalue table, see S7 Table). The LM test in S8 Table shows that we cannot reject the null hypothesis that there is no serial correlation, and the Jarque-Bera tests in S9 Table show that we cannot reject that the residual series of the liq_t equation is normally distributed, so residual series of the liq_t equation in the VAR is Gaussian white noise.

The SVAR writes where vector Y_t = [slf_t, liq_t]′, A is the short-run restriction matrix. There are k = 2 endogenous variables in the model. We need the k × (k − 1)/2 = 1 restriction to identify the model. Because Shibor is not market benchmark rate in China, we suppose that in the short run, Shibor has no impact on SLF, while SLF has an impact on Shibor, which indicates that the A matrix is written as where element a is a parameter to be estimated. The Granger causality test (see S10 Table) supports our restriction setting. Vector X_t = [gdp_t, omo_t, reserve_t]′ is the exogenous variable vector. P₁ and P₂ are parameter matrices. Vector e_t denotes the error term. We believe that it is reasonable here to suppose that different monetary policy tools are used independently; moreover, the time span of our study is much shorter than the policy time lag, so monetary policy tools other than SLF and economic growth work as exogenous control variables.

The orthogonalized impulse response function of the interbank rate to the impact of SLF expansion is plotted as Fig 3, and the 90% confidence interval is shown by the shaded area. We can observe that the instant impact of SLF on the interbank rate is raising the interbank rate. The result confirms the existence of the anomaly formally; the expansion of SLF quantity raises the interbank rate, not cuts it.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Interbank rate impulse response to SLF.

https://doi.org/10.1371/journal.pone.0284470.g003

Transmission channel analysis

To explain the anomaly, we take two steps. The first is to verify the risk-taking channel of SLF with Eq (1). The result in S11 Table is summarized in Table 2, where insig. means an insignificant parameter.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Monetary policy effect on risk-taking.

https://doi.org/10.1371/journal.pone.0284470.t002

From the Overall column, we can observe that SLF quantity expansion spurs bank risk-taking activity generally, which verifies hypothesis (1). Specifically, every 10 trillion CNY increase of the outstanding SLF quantity adds 0.001 points to the bank z-score. This result aligns with those of some other studies, indicating the effectiveness of bank risk-taking channel. For example, with bank lending standards data in the Euro zone and the US [19], shows that a lower short-term policy rate tends to soften bank lending standards and stimulate the expansion of financial institution balance sheet. This effect is amplified when there is a high level of financial innovations. Their study shows that an increase in the overnight rate significantly soften banks standards for corporate loans, mortgage loans, and consumer loans. Similarly [22], examines the connection between monetary policy and corporate lending standards in the US with data from Senior Loan Officers Opinion Survey, and it verifies the greater appetite of banks in periods of an easy monetary policy. With credit register data in Spain [21], arrives at a similar conclusion: The relatively low short-term rate prior to the loan tends to encourage banks to lend to borrowers with bad or no credit history. These papers detail the risk-taking effects in some industrial economies. For the developing country case [20], studies the bank risk-taking behavior in the credit market in a dollarized country, Bolivia. Their study shows that a reduction of 100 basis points in the funds rate increases the likelihood of a borrower with non-performing loans getting approved for a loan by 1.1 percentage points. As for China [40], uses the non-performing ratio as the proxy of bank risk-taking behavior, and M2 as the proxy of the monetary policy stance. The regression result indicates the effectiveness of the bank risk-taking channel of money stock (M2). These studies provide strong empirical evidence for the theory of risk-taking. However, the risk-taking effect on the Chinese interbank market is under-researched. Our work fills this gap. And our regression results are consistent with the literature, namely, that monetary policy (in this case, SLF) has an encouraging effect on banks’ risk-taking. In other words, loose SLF policy may lead to an increase in money demand, which in turn may result in liquidity tension.

The columns NSOCB and SOCB present estimations with NSOCB and SOCB data, respectively. These columns indicate the differential reactions of NSOCB and SOCB to SLF. The parameter for NSOCB is insignificant, and that for SOCB is significantly positive. The regressions above mean that the risk-taking encouragement of SLF is greater for SOCBs. As for the differential impact on heterogeneous banks, the following heterogeneity analysis section examines that in depth.

Regression (1) has verified the existence of the bank risk-taking channel in the Chinese interbank market, but this does not necessarily mean that the risk-taking effect will dominate the liquidity effect of SLF expansion; in other words, the interbank market rate does not necessarily increase. So secondly, we need to verify hypothesis (2) by the regression Eq (2). Similarly to regression (1), regression (2) is conducted with SOCB data, NSOCB data, and all of the data. The result (see S12 Table) is summarized in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Mediation effect of risk-taking.

https://doi.org/10.1371/journal.pone.0284470.t003

Parameters of the z-score for all three regressions are significantly positive. This result strongly supports hypothesis (2). The risk-taking activity of banks raises the interbank bid rate. Specifically, every point increase in the z-score adds an average of 448.253 percentage points to the bank interbank market bids. In other words, every 10 trillion CNY increase in the outstanding SLF quantity adds about 0.45 percentage points on average to interbank market bids. According to the risk-taking channel theory, we can explain the basic mechanism of the whole process. There is a substitution of interbank lending and commercial loan. Owing to the arbitrage-free principle, the expected yield for interbank lending and commercial loans should be equal. The interest rate for commercial lending is determined by the marginal productivity of capital, which is not determined by SLF in the short run. The probability of default for corporate loans is normally higher than that for interbank lending. Additionally, it is reasonable to assume that the expansion of SLF has a more significant effect on reducing the probability of default for corporate loans than on reducing the probability of default for interbank lending. Thus, the interbank lending rate has an upward trend. Meanwhile, the liquidity effect causes the interbank rate to exhibit a downward trend. Our regression unveils that the upward trend beats the downward one, and interbank rate increases.

Most literature on bank risk-taking channel refers to the above study on the effect of monetary policy on lending standards, credit availability, or credit cost. Some others study on the connection between bank characteristics [41], bank competition [42], corporate governance [43], and their risk-taking behavior. Very few compare the relative importance of the risk-taking and liquidity effects of monetary policy on the interbank market. [16] argues that liquidity effect dominates the risk-taking effect. However, the results in regression (2) reveal that in the Chinese interbank market, the risk-taking encouragement effect is greater than the liquidity effect of SLF.

Our transmission mechanism analysis actually combines the literature on the macro effects of monetary policy and the heterogeneous risk-taking behavior of banks. Literature examining the relationship between monetary policy and macroeconomic variables often uses time series data, which contain less information than panel data. On the other hand, literature studying bank risk-taking behavior tends to use panel data, but few have examined the macro effects of monetary policy. This subsection combines the two strands of literature, and emphasizes the importance of SLF as a tool for expectation management and forward guidance, not simply as a tool for liquidity management.

Heterogeneous analysis

From the regressions above, we notice that though risk-taking channel theory generally works, so the ownership factor strengthens or weakens the risk-taking effect of SLF. In this part, we consider more observable bank heterogeneity factors, and explore how they affect bank risk-taking behaviors. By regression (1), we can verify the inference of hypothesis (3). We find that some certain interaction term parameters are significant, which supports the inference and hence hypothesis (3).

Now, we focus on the following question: How do any of the three heterogenous transmission theories mentioned above (CRT, MFT and IIT) work? The parameters of SLF interaction terms (see Table 4) are summarized in S13 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Bank accounting indices effect.

https://doi.org/10.1371/journal.pone.0284470.t004

According to Table 1, we can tell the effectiveness of CRT, MFT, and IIT for NSOCB and NSOCB, and we observe the accounting indices through which these theories affect the interbank rate. The result is presented in Table 5.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Bank heterogeneity.

https://doi.org/10.1371/journal.pone.0284470.t005

CRT and MFT are supported by all of the sample sets, while IIT is not supported by the NOSB data. The effectiveness of CRT means that banks with better accounting indices are less costly in equity capital financing. Thus, they can lower their interbank liquidity bids more flexibly as SLF quantity expands. By contrast, the reaction of the relatively worse-capitalized banks is less active. Both the overall and NSOCB sample sets show that the main accounting index that transmits CRT is capital adequacy. This implies that capital adequacy regulation can help suppress the risk-taking impulse of worse-capitalized banks and reduce liquidity risk. For SOCBs only, CRT works via the asset growth rate. This means that although potential shareholders of banks generally care more about the owner’s equity (capital) when they are making their investment decisions, those of SOCBs have more confidence in the banks’ risk management capability and thus care more about their profitability.

NPL is the main accounting index that transmits MFT for all of the sample sets. This indicates that, based on traditional MFT literature [33], we find that NPL is the determinant of the reserve financing cost. Different from most equity investors, creditors (including other banks) care more about the asset side of the balance sheet. Operating income growth rate transmits MFT for only SOCBs.

IIT works for the overall sample set. Specifically, it works better for NSOCBs. NSOCBs in China are more reliant on the net interest income [44], so NSOCBs as a group tend to be affected by the net interest income to a larger extent.

The heterogeneity analysis is a beneficial supplement to the transmission channel analysis. We find that CRT, MFT, and IIT theories [29, 33, 34] operate through different factors. SOCB and NSOCB face differential challenges in the capital market and reserve market. A proper SLF policy design should consider market situations. Additionally, this result can be useful for bank liquidity management.

Robustness

We mainly perform two robustness tests. First we consider bank heterogeneity using a dynamic panel data model: (4)

The significance of parameter δ determines whether SLF quantity expansion impacts the interbank market rate, and the sign determines the direction of the impact.

Considering the liquidity stratification of the banking system, regression (4) is conducted with not only the whole data set but also the NSOCB and SOCB data. The result (see S14 Table) is summarized in Table 6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. SLF effect with panel data.

https://doi.org/10.1371/journal.pone.0284470.t006

The result of regression (4) agrees with that of the SVAR model above. The expansion of the outstanding SLF quantity drives up the interbank market rate. In addition, it shows that interbank bid quotes of both SOCB and NSOCB are positively driven by the outstanding SLF quantity. Specifically, according to the regression result, for each trillion CNY increase of the outstanding SLF quantity, the NSOCB interbank lending bid will increase by 1.03 percentage points, and the SOCB interbank lending bid will increases by about 1.01 percentage points. As we have analyzed in the heterogeneity part, the inclusion of bank heterogeneity does not change the main conclusion of our observation; it only adjusts the extent to which banks are affected by SLF policy. This result shows the robustness of the risk-taking channel theory in terms of bank heterogeneity.

Second, we substitute the 1-month Shibor (r_it and r_it−1) in Eq (4) with overnight, 1-week, 2-week, and 3-month Shibor respectively. Some papers show the different risk-taking effects of monetary policy tools. For instance, [19] finds that the short-term policy rate has a greater effect than the long-term rate. However, few works examine the different risk-taking effects of monetary policy on the interbank market rate with different maturities. This is what we do in the second robustness test. The result (see S15 Table) is summarized in Table 7. The regression results obtained using Shibor data with different maturities are all consistent with the theory of bank risk-taking channels. Specifically, SLF policy adjustment has a greater effect on Shibor with longer maturity. Interbank rates with a longer maturity are normally believed to have greater risks, so the amplification of the expectation effect is greater. This result shows the robustness of the risk-taking channel theory in terms of maturity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Shibor with different maturities.

https://doi.org/10.1371/journal.pone.0284470.t007

Conclusively, our analysis on the risk-taking channel of SLF is robust in terms of both bank heterogeneity and interbank rate maturity. This indicates that the results of this paper can be used to analyze the SLF influence on banks with different financial features. Furthermore, the conclusions can be extended to interbank markets with different maturities.