Collateral monetary policy, shadow banking and bank risk: evidence from China Free

Stringent financial regulations and regulatory arbitrage contribute to the growth of the global shadow banking system (Gennaioli et al., 2013; Plantin, 2015; Zhu, 2021). Various financial regulations prohibit banks from autonomously creating private money (deposits) through asset expansion (Werner, 2014; Jakab and Kumhof, 2015). The fostered unregulated financial activities caused by the tightening of financial regulations affect the banking system stability and disrupt monetary policy transmission, which has attracted huge attention (Bengtsson, 2013; Le et al., 2022).

However, the role of collateral monetary policy ^[1] in contributing to the shadow banking system remains an underexplored area. Unlike financial regulation, collateral monetary policy operates through a distinct mechanism but achieves similar outcomes in influencing shadow banking activities via bank’s money creation behaviors. The acquisition and retention of an adequate monetary base enable banks to expand loans, comply with financial regulations and serve as the initial steps in monetary policy transmission (Yuan et al., 2024). So far, limited attention has been paid to the impact of monetary base distribution under the collateral monetary policy on the shadow banking system and bank risk. This underscores the need for further research into the shadow banking system and bank risk, with a specific focus on the monetary base distribution of collateral monetary policy.

Collateral monetary policy matters due to its discrimination in the allocation of monetary base among banks (Nyborg, 2017; Yuan et al., 2024). In most countries, central banks have injected extensive monetary base into the banking system through various collateral monetary policy. However, concerns arise regarding discrimination against heterogeneous financial institutions. Only qualified financial institutions can obtain central bank liquidity by pledging designated qualified collateral (Andrade et al., 2019). This pledging not only incentivizes banks to pledge assets that meet collateral eligibility criteria to secure central bank liquidity (Fang et al., 2020) but also biases liquidity provision, exacerbating lending inefficiencies and destabilizing the banking system (Nyborg, 2017). For instance, in the Eurozone, primary banks with direct access to European Central Bank (ECB) liquidity have strong market power over non-primary banks, disrupting the transmission of monetary policy (Eisenschmidt et al., 2022). This paper, focusing on China, seeks to deepen the broader understanding of collateral monetary policy and its implications.

The heterogeneity in bank liquidity acquisition caused by the collateral monetary policy compels non-primary banks to engage in the shadow banking system. Primary banks in China —predominantly large banks— benefit from abundant monetary bases from various central bank liquidity instruments (Li, 2021). Conversely, most non-primary banks face significant constraints in accessing liquidity from the People’s Bank of China (PBOC) due to insufficient holdings of qualified collateral (Sun et al., 2021). Therefore, the non-primary banks resort to issuing wealth management products (WMPs), the major component of the shadow banking system in China, to complete household deposits and interbank funding, thereby generating corresponding risks (Shao et al., 2020). These WMPs which offer higher yields than deposit rates enable non-primary banks to finance firms and circumvent regulations (Hachem and Song, 2016). Consistent with Chen et al. (2018), non-primary banks exhibit heightened reliance on WMPs during periods of monetary tightening. The heterogeneity in liquidity distribution towards bank also undermines the efficiency of monetary policy transmission (Jiang et al., 2023). Moreover, the discrimination of collateral monetary policy contributes to the prosperity of shadow banking and increases bank risk.

This paper aims to explore the interlinkages among collateral monetary policy, shadow banking system and bank risks in China. The existing literature overlooked the response of the shadow banking system to central bank collateral monetary policy over the past decade. The shadow banking system, an integral component of contemporary financial structures, has been criticized for exacerbating financial leverage, undermining financial market stability and diminishing the effectiveness of monetary policy (Yang et al., 2019). This paper underscores the distorting effect of collateral monetary policy on the expansion of the shadow banking system. By unveiling the relationship between collateral monetary policy, shadow banking system and bank risks, this paper contributes to advancing the central banks’ liquidity management framework. Previous studies identify that flourishing shadow banking activities are accountable for increased bank balance-sheet risk, ineffectiveness of monetary policy and accumulated systemic risk (Pellegrini et al., 2022). Building on this foundation, this paper delves deeper into how the shadow banking system amplifies bank risks by facilitating the uncontrolled creation of money following the implementation of collateral monetary policy.

This paper makes four distinctive contributions to the literature. First, this paper fills the research gap by exploring the distorting effect of collateral monetary policy on liquidity distribution and the shadow banking system, which have received limited attention thus far. By highlighting the heterogeneity in monetary base acquisition in China and unveiling the money creation mechanisms underlying the shadow banking system, this paper presents the stimulative effect of collateral monetary policy on the shadow banking system. Second, it extends the understanding of bank risk by quantifying the responses of bank risk, derived from the NCDs market, to collateral monetary policy. Third, this paper employs the SHAP-XGBoost method to identify and attribute the role of collateral monetary policy in the shadow banking system and bank risk. Fourth, the study investigates how the relationship among collateral monetary policy, shadow banking system and bank risk evolves before and after the implementation of China’s 2018 New Asset Management Regulations (NAM Regulation). By leveraging machine-learning insights, it enhances the understanding of the impact of financial regulation on these dynamics.

The remainder of this paper is organized as follows. Section 2 presents the conceptual framework to outline the link between collateral monetary policy, shadow banking system and bank risk. Section 3 provides an overview of collateral monetary policy within China’s tiered banking system, shadow banking system and associated bank risk and proposes the hypotheses. Section 4 and Section 5 detail the research design and present empirical results. Section 6 discusses the broader implications of the results, linking them to the existing literature, while Section 7 concludes the paper.

This section is comprised of three parts. The first demonstrates how banks provide financial services and create money through traditional loans. The second clarifies how the shadow banking system assists banks to evade the financial regulations and constraints. The third establishes the connection among collateral monetary policy, shadow banking system and bank risks. It is worth emphasizing that certain equations are employed to effectively illustrate financial regulations and constraints faced by banks rather than for equilibrium analysis. That is, this paper does not seek to construct a general equilibrium model. Additionally, tables describing the changes in bank assets and liabilities are used to link shifts in regulatory indicators with the weights assigned to assets and liabilities by regulators, providing a comprehensive view of the regulatory framework.

Shadow banking activities create money through the “interbank lending” channel and WMPs (Table 1), as pointed by Sun (2019). Supposing that commercial bank C is unable to directly finance borrower E due to various financial regulations, it opts to lend money to non-financial institution D with a high credit rating. Correspondingly, bank C would concurrently hold X amount of funds in “Inter-bank Lending or Lending to non-bank financial institution D” (asset items) and “deposit of D” (liability items). Subsequently, financial institution D transfers the funds to borrower E, the owner of a bank account in bank C. This results in a decrease of X amount of money in the deposit of D and an equivalent increase of X funds in the deposit of E. Thus, bank C creates money (deposits of E) by holding high-quality claims to financial institution D (asset), rather than the loan of borrower E.

Table 2 illustrates how guaranteed and non-guaranteed WMPs create money. The objective of commercial bank C is to finance borrower E by issuing WMPs. Assuming investor F purchases guaranteed WMPs with a value of X with the investor’s deposit account in bank C, the investor’s deposits are immediately transformed into structure deposits, indicating investor F’s holdings of guarantee WMPs. Then, investor F’s deposits are transferred to borrower E’s account in bank C. Correspondingly, bank C holds X amount of specific assets (asset) and X funds of borrower E’s deposit (liability) concurrently. From the alterations of the balance sheet of bank C, it can be discerned that money (deposits of E) is created by holding specific assets rather than issuing loans to borrower E.

The key distinction between guaranteed and non-guaranteed WMPs lies in the fact that bank C removes the “specific assets with X value” and “deposits of the investor F with X value” from its balance sheet. Nevertheless, these shadow banking activities result in the creation of money. Our analysis aligns with the empirical findings of Chen et al. (2018) who highlighted that the shadow banking system mitigates the effect of traditional monetary policy on bank credit.

The costs of issuing WMPs and structure deposits typically surpass those of conventional deposits in most cases. So, why does commercial bank C engage in shadow banking activities with higher interest expenses? This question warrants further exploration from the perspectives of financial regulations and collateral monetary policy.

Previous studies have primarily focused on the loan-to-deposit ratio (LDR) cap imposed within banking systems (Le et al., 2022). The effectiveness of LDR cap is fundamentally insufficient, as banks hold non-loan assets and create money (deposits) through various shadow banking activities, which facilitates banks circumventing the LDR constraints. Banks are also subject to various financial regulation items, including capital regulation, liquidity regulation and required reserve ratio etc. The definition of capital regulation is expressed in equation (1):

where $CAR$ is the bank’s capital adequacy ratio and $CARmin$ is the minimum required capital adequacy ratio. $K$ and $RWA$ present capital and risk-weighted assets, respectively. The definition of $RWA$ is presented in equation (2):

The composition of risk-weighted assets encompasses traditional loans $L$⁠, other assets $OA$ (including interbank lending and specific assets detailed in Tables 1 and 2 and off-balance assets $Loff$ with a discount $ξ$⁠. The weights of $L$⁠, $OA$ and $Loff$ are $α1$⁠, $α2$ and $α3$⁠, respectively. Therefore, the capital regulation is defined in equation (3):

As discussed in Section 3.1, the quality of loans from financial institution D and specific assets outweighs that of borrower E. Consequently, the risk weight $α2$ is lower than $α1$⁠. Similarly, the interactive term of risk weight $α3$ and $ξ$ is lower than $α1$⁠. Despite the higher costs associated with engaging in the shadow banking system compared to traditional loans, this system enables banks to finance a multitude of borrowers E while minimizing the consumption of capital. In addition, the required reserve ratio and net stable funding ratio are also served as significant constraints for banks. Table 3 presents a typical bank C’s balance sheet. Thus, undercapitalized banks are significantly more inclined to engage in shadow banking activities to circumvent financial regulations, suggesting that shadow banking activities pose substantial risks to the banking sector (Zhu et al., 2019).

The required reserve ratio and liquidity regulation of a typical bank C should be satisfied by the following equation:

where $τ$ is the required reserve ratio, $Rmin$ is the required reserve, $NSFRmin$ is the required net stable funding ratio. According to the regulations of the China Banking and Insurance Regulatory Commission (CBIRC), the weights of assets and liabilities in NSFR are listed in Table 4.

According to the money creation process in Tables 1 and 2, the non-guaranteed WMPs facilitate banks in removing the deposits from bank balance sheets, thereby alleviating the constraints imposed by the required reserve ratio. Additionally, the weight of loans consistently outweighs that of interbank lending, as well as specific and off-balance assets. Consequently, banks can finance firms with lower quality ratings via the shadow banking system while adhering to liquidity regulations, albeit at the expense of higher costs (equation 5) (Hachem, 2018).

So far, we have demonstrated how a shadow banking system enables banks to circumvent financial regulations. Subsequently, we introduce the primary driving force, pertaining to collateral monetary policy, which positions numerous small- and medium-sized banks at a disadvantaged position in securing liquidity from PBOC.

Since 2014, the PBOC has relied on collateral monetary policy instruments to inject a monetary base into the banking system, which supports primary banks in advance, leaving non-primary banks to obtain a monetary base from primary banks in the interbank market (Yuan et al., 2024). This heterogeneity between primary and non-primary banks gives rise to a tiered banking system in China (Jiang et al., 2023). Consequently, when primary banks require a monetary base to satisfy the required reserve ratio and liquidity requirements, they can apply for liquidity from the central bank. However, due to the lack of access to liquidity from the central bank, non-primary banks are compelled to engage in shadow banking systems to finance more borrowers and comply with various regulations. Recent evidence has revealed that the bias of collateral monetary policy incentivizes small- and medium-sized banks to issue more WMPs, which are the primary components of the shadow banking system in China (Shao et al., 2020).

Overall, banks are subject to various financial regulations that restrict money creation. Primary banks, capable of procuring ample liquidity from the central bank, can fulfill these regulatory requirements at a minimal cost. However, non-primary banks, unable to directly access liquidity from the central bank, rely on the shadow banking system to alleviate the regulation constraints and satisfy the financial demands of borrowers.

In this section, we establish the connection between collateral monetary policy, shadow banking system and bank risks. The monetary base distribution framework of collateral monetary policy, which disproportionately allocates monetary base to primary banks, creates a scarcity of liquidity for non-primary banks. Consequently, non-primary banks are compelled to turn to the shadow banking system to finance their clients, thereby circumventing regulatory constraints related to an insufficient monetary base. This reliance on shadow banking exposes banks to risks transferred from borrowers within the shadow banking system and amplifies risk-taking through regulatory arbitrage and increased leverage (Gennaioli et al., 2013).

Previous studies highlight that banks engage in shadow banking activities primarily to finance high-risk sectors, including the real estate market, local government financing platforms and over-leveraged non-state-owned enterprises with excess capacity. These borrowers, characterized by elevated default probabilities and severe maturity mismatches, pose significant risks to the banking system (Bleck and Liu, 2018; Chen et al., 2018; Allen et al., 2019). Moreover, banks often provide implicit guarantees and rigid payment for shadow banking products, exposing themselves to potential losses when defaults occur. This implicit risk transfer from investors to banks exacerbates banks' exposure to financial losses, increasing bank risks (Huang et al., 2023).

Given that banks effectively create money and generate non-loan assets with risk profiles comparable to loans by participating in the shadow banking system, they should maintain sufficient capital and liquidity to absorb potential losses and maintain liquidity shocks. However, the absence of effective supervision leads to inadequate provisioning of these buffers. This regulatory gap exacerbates systemic fragility and increases bank risks. The shadow banking activities imply the overestimation of the regulatory capital adequacy ratio and liquidity ratio as the denominator of risk-weighted assets is significantly underestimated (Wu and Shen, 2019). Consequently, a thriving shadow banking system signals greater risk exposure for banks and a more vulnerable banking sector, overall (Huang and Shen, 2019) (see Figure 1).

China’s foreign exchange reserves peaked at over $2.7 trillion in June 2014 before entering a period of decline (Figure 2). To mitigate the shocks of foreign exchange outflows which contracted the monetary base, the PBOC employed open market operations (OMOs) and introduced various lending facilities, including Standing Lending Facilities (SLOs), Short-term Lending Facilities (SLFs) and Medium-term Lending Facilities (MLFs), to inject monetary base into the banking system (Zhang and Tan, 2015; Jiang et al., 2023). All these instruments require banks to submit qualified assets in exchange for a monetary base from the PBOC. Consequently, there was a rapid expansion in both claims on depository corporations (CODC) as well as the stock of MLFs on the PBOC’s balance sheet (Yuan et al., 2024).

Table 5 compares the characteristics of collateral monetary policy in China and the USA. The Federal Reserve System (FED) has developed a range of monetary policy instruments to provide liquidity of varying maturities to diverse financial institutions, accepting a broad spectrum of collateral types (Li, 2021). In contrast, collateral monetary policy instruments established by the PBOC distribute the majority of liquidity to primary banks, creating significant disparities in liquidity access across different banks (Shao et al., 2020). Furthermore, high-quality collateral is predominantly held by primary banks, exacerbating the challenges faced by non-primary banks in securing liquidity (Sun et al., 2021).

The interbank market plays a critical role in the funding needs of non-primary banks (Allen et al., 2009). As shown in Figure 3, primary banks supply excessive liquidity (monetary base) to non-primary banks in the interbank market. However, the interbank market primarily facilitates the transmission of short-term liquidity and rendering it less suitable for addressing long-term liquidity needs (Jiang et al., 2023). While non-primary banks can recycle short-term liquidity to finance long-term projects, this strategy incurs higher funding costs and exposes them to significant refinancing risks. In contrast, the shadow banking system offers a more compelling alternative for non-primary banks through three channels. First, shadow banking products provided by non-primary banks typically yield higher returns than traditional deposits, enabling these banks to compete for household and corporate deposits. These products not only supplement their liquidity on the liability side but also help meet regulatory requirements. Second, shadow banking products facilitate the exchange of long-term liquidity between primary and non-primary banks, effectively bypassing the limitations of the interbank market and circumventing regulatory constraints (Hachem, 2018). Third, by offering financing to non-financial firms through shadow banking products, non-primary banks can circumvent numerous regulations and government lending guidelines, albeit at a higher cost compared to traditional loans (Bleck and Liu, 2018). Thus, we propose the first hypothesis:

Moreover, considering that the goal of shadow banking activities is to provide financing services to firms, these products create money in various forms. While Sun (2019) focuses on commercial banks’ balance sheet adjustments, this paper extends the analysis of interlinkages among shadow banking, collateral monetary policy and bank risk by offering a more nuanced view of these adjustments. Since banks ultimately bear the potential losses arising from the default events of shadow banking borrowers, these activities essentially function as high-risk loans. Furthermore, by circumventing financial regulation, banks fail to maintain adequate capital and liquidity buffers to absorb potential losses, thereby increasing their leverage and exposing themselves to risks that are not fully reflected in publicly disclosed balance sheet data. Accordingly, the second hypothesis is proposed:

To investigate the two hypotheses, this paper first empirically examines the role of collateral monetary policy in driving the expansion of the shadow banking system. Thereupon, it explores the contributions of collateral monetary policy and shadow banking system to bank risks using the SHAP-Bayesian-XGBoost machine-learning method.

The dataset used spans from January 2015 to December 2023. All relevant data can be accessed via the Wind database and bank annual reports. To mitigate the effects of seasonality, all time-series data are seasonally adjusted using the X-11 method. Before the empirical test, we remove individual fixed effects for all bank-level variables to enhance the robustness of results.

When the PBOC injects or withdraws monetary base (a flow variable) to the banking system by collateral monetary policy instruments, it alters the amount of monetary base held by banks, thereby influencing bank credit expanding behaviors. However, solely considering the monetary base injected or withdrawn by the PBOC within a single year does not accurately capture the impact of these instruments as it overlooks the volume of the existing monetary base. Therefore, the “ratio of monetary base injected by the collateral monetary policy in year t to total monetary base” is a more appropriate indicator, despite its nature as a ratio between flow and stock variables.

Three variables are used to measure the collateral monetary policy. First, the index “ratio of monetary base injected by MLF to total monetary base in circulation” (MLFR) gauges the importance of MLF in the broader banking system. The second variable is “the ratio of the claims to deposit companies to total monetary base” (CODCR), which servers as a proxy to evaluate the overall monetary base injected by all the collateral monetary policy instruments. Third, this paper measures “the ratio of monetary base that injected by the non-MLF instruments to total monetary base” (non-MLFR) by subtracting the MLFR from the CODCR. Therefore, the paper examines the effect of MLF, the overall collateral monetary policy and non-MLF instruments on both the shadow banking system and bank risk.

The measurement of the shadow banking system (SB) and its role in money creation has been addressed by Sun (2019) and Li (2019). In this paper, we estimate the volume of the shadow banking system by considering the liabilities and assets on banks' balance sheets, as detailed in Table 6. The equity item is excluded as we focus exclusively on the direct correspondence between shadow assets and money. Figure 4 illustrates that the shadow banking volume has grown significantly between 2010 and 2017. However, when the NAM Regulation in 2018 was introduced, it turned downward.

This paper primarily measures bank risk by trading data from the NCDs market where professional financial institutions are primary investors. NCDs are bonds issued by commercial banks to raise wholesale funding in the interbank market. The yield to maturity (YTM) of NCDs reflects the fund cost of banks. Additionally, the credit spread measured by the difference between the YTM of NCDs and risk-free rates, serves as an important indicator of bank risk. Investors in China’s interbank market grasp comprehensive and timely information about banks, allowing the credit spread of NCDs to offer a more accurate and immediate reflection of bank risk compared to traditional risk indicators, such as Z-scores or risk-weighted asset ratios, which are based on annual financial statements. The effectiveness of these traditional risk variables is often diminished by the opacity introduced by shadow banking activities, which reduces the information content of indicators like capital levels. Therefore, this paper uses the difference of the NCD’s TYM and 1-year treasury bond yield as the proxy of bank risk.

In investigating the driving effect of collateral monetary policy on the shadow banking system, this paper also selects a set of macroeconomic variables, as detailed in Table 7. For the analysis of bank risks, a series of both macroeconomic and bank-level variables are selected, which are also presented in Table 7. The definitions of these variables are provided in the same table for clarity.

XGBoost is an ensemble model built on an efficient implementation of decision trees, designed to produce a combined model with superior predictive performance (Jabeur et al., 2024). To enhance model performance and address overfitting concerns, three key hyperparameters are optimized: the number of trees (n_estimators), the maximum tree depth (max_depth) and the learning rate (learning_rate). These parameters are fine-tuned using the Bayesian Optimization library, which systematically identifies optimal values (Shi et al., 2021).

To interpret results and uncover the “black box” of machine learning in analyzing the relationships between collateral monetary policy, shadow banking system and bank risks, Shapley Additive Explanations (SHAP) are employed (Aas et al., 2021). SHAP provide robust advantages for dynamic analysis, enabling a deeper understanding of how the influence of various factors evolves over time.

This section presents empirical results of the relationship between collateral monetary policy, shadow banking system and bank risks. Empirical results first present the overall stimulating effect of collateral monetary policy on the shadow banking system. Then, this paper further investigates the role of collateral monetary policy and shadow banking system in bank risk. In addition, this section highlights how these relationships evolved before and after the implementation of China’s NAR Regulations in 2018. The descriptive statistics are presented in Table 8.

This paper employs four machine-learning methods -Support Vector Regression (SVR), Random Forest, Gradient Boosting Decision Trees (GBDT) and XGBoost-to fit the data and evaluate model performance using R-square, mean squared error, explained variance score and mean absolute error (Table 9). The results, summarized in Table 9, indicate that XGBoost outperforms the other methods across all evaluation scores. To optimize the XGBoost model, Bayesian hyperparameter tuning is conducted using the Bayesian Optimization library. The optimal parameters are determined to be 359 for the number of trees (n_estimators), 3 for the maximum tree depth (max_depth) and 0.1 for the learning rate (learning_rate).

Figure 5 highlights two key observations regarding the contribution rankings of variables influencing the shadow banking system. First, variables related to collateral monetary policy emerge as dominant drivers with nonMLFR ranking 1st, CODCR 2nd and MLFR 11th, underscoring their central role in shaping the composition of the shadow banking system. Second, FER ranks as the 5th most significant contributor, indicating that shifts in the monetary base injection mechanism are pivotal factors in the development of the shadow banking system. These findings align with the analysis presented in Section 2.2 and provide empirical support for Hypothesis 1.

This paper further explores the non-linear relationship between collateral monetary policy variables (CODCR, MLFR and nonMLFR) and shadow banking system (SBR), as illustrated in Figure 6. The results reveal a dichotomy. Specifically, when CODCR is below 35, it exhibits a positive relationship with SBR, suggesting that collateral monetary policy stimulates the shadow banking system, thereby increasing the ratio of money created by shadow banking activities to M2. This reflects a period during which banks, particularly those with limited access to central bank liquidity, utilized the shadow banking system to circumvent financial regulations and meet borrowers' financing needs, as discussed in Section 2. In contrast, when CODCR exceeds 35, its relationship with SBR turns negative. Notably, such instances occurred exclusively after June 2018, following the introduction of the NAM Regulations in April 2018. This shift implies that the new regulations curtailed commercial banks' shadow banking activities, effectively neutralizing the stimulating effects of collateral monetary policy on the shadow banking system. Similar patterns are observed in the relationships between MLFR, nonMLFR and SBR, further reinforcing the significant impact of the 2018 NAM Regulations in China. These findings provide additional empirical support for the regulatory measures' effectiveness in reshaping the dynamics of shadow banking.

To ensure robustness, this paper individually inputs the three collateral monetary policy variables into the SHAP-XGBoost model to test their respective driving effects on the shadow banking system. The results presented in  Appendix remain consistent, reinforcing the validity of our findings.

This section explores the role of collateral monetary policy and the shadow banking system in shaping bank risks. Figure 7 ranks each variable’s contribution to bank risk, highlighting that SBR is the most significant driver, while collateral monetary policy variables rank third in their impact. Compared to other macroeconomic and bank-level variables, collateral monetary policy and shadow banking system emerge as critical factors deserving greater attention. These findings underscore the importance of regulators dynamically adjusting their supervisory strategies to align bank regulation intensity with the implementation of collateral monetary policy. Figure 8 provides further insights into the non-linear effects of collateral monetary policy and shadow banking system on bank risk, as well as the impact of the 2018 NAM Regulations. As shown in Figure 2, the values of CODCR, MLFR and nonMLFR steadily increased, surpassing thresholds of 0.34, 0.13 and 0.19, respectively, after April 2018. Figure 8(a) reveals that before the 2018 NAM Regulations, CODCR, MLFR and nonMLFR exhibited a positive relationship with bank risks, underscoring the significance of the monetary base injection mechanism in risk formation. Before the regulator forbade the channels as discussed in Section 2, banks that lack of monetary base from the central bank will rush into shadow banking activities to provide financing for borrowers and generate corresponding money. However, the financial assets and money created by the shadow banking system are out of financial regulations and hide the potential risks transferred to banks. As a result, financial institutions with information advantages require higher credit spreads when purchasing NCDs to compensate for the potential risks associated with banks’ involvement in the shadow banking system.

The NAM Regulations enacted in April 2018 fundamentally altered this relationship. Engaging in shadow banking no longer provided opportunities for regulatory arbitrage. For example, the risk-weighted asset (RWA) weight for off-balance-sheet assets rose from 5% before the regulations to 100%, significantly narrowing the gap between off-balance-sheet assets and traditional loans. The NAM Regulations effectively closed loopholes, curtailing banks’ ability to circumvent regulations and mitigating risks previously associated with shadow banking activities.

Figure 8(d), (e) and (f) illustrate the non-linear relationship between the shadow banking ratio (SBR) and bank risk. During the sample period from January 2015 to December 2023, SBR values below 27.6 were observed exclusively after June 2018, while values exceeding 27.6 occurred only before that date. This distinction highlights a significant shift in the risk associated with shadow banking activities. Before June 2018, banks engaging in shadow banking activities to finance borrowers faced heightened risks. These institutions were often inadequately capitalized and lacked sufficient liquidity buffers to absorb potential losses arising from borrower defaults. As a result, the increase in SBR during this period contributed directly to elevated bank risks. However, the introduction of the 2018 NAM Regulations fundamentally altered this landscape. By mandating that banks increase the risk weight for off-balance-sheet activities from 5% to 100%, the regulations require institutions to hold adequate capital and liquidity to absorb potential losses. The new regulation effectively neutralized the risk transmission channel through which shadow banking previously contributed. Borrower-originated risks are now absorbed within banks’ capital buffers and liquidity buffers, mitigating their impact on overall bank risk levels. As a result, post-regulation, the shadow banking system no longer poses a significant risk to banks, underscoring the effectiveness of the 2018 regulation in addressing regulatory arbitrage and reinforcing financial stability.

This paper further explores the heterogeneous responses of bank risk to collateral monetary policy and the shadow banking system, highlighting significant disparities between primary and non-primary banks in their access to central bank liquidity. In Figure 9, results show that collateral monetary policy and shadow banking system are still major drivers of bank risks for both primary and non-primary banks. In addition, the ranks of collateral monetary policy variables in non-primary banks are higher than that in primary banks, which indicates that non-primary banks are more sensitive and affected by collateral monetary policy.

The effects of collateral monetary policy on non-primary banks are depicted in Figure 10(a), (c), and (e), while the corresponding effects on primary banks are shown in Figure 10(b), (d), and (f). Before the implementation of the NAM Regulations, the MLFR was positively associated with bank risks for non-primary banks but negatively associated with bank risks for primary banks. Non-primary banks, unable to secure sufficient liquidity through the MLF, were driven to engage in shadow banking activities, thereby elevating their risk levels. Conversely, primary banks, benefiting from access to medium-term liquidity provided by the central bank, stabilized their funding structures and reduced risk.

Following the introduction of the NAM Regulations, the MLF mitigated bank risks for both primary and non-primary banks. However, the impact of non-MLF collateral monetary policy varied. Before the NAM Regulations, nonMLFR increased risks for non-primary banks while reducing risks for primary banks. After the NAM Regulations, non-MLF policies began to reduce risks for non-primary banks while having minimal impact on primary banks. This shift reflects a partial redirection of liquidity from non-MLF instruments to support non-primary banks.

This paper undertakes four robustness checks ( Appendix) to validate the empirical results. First, the machine-learning method is switched to Gradient Boosting Decision Trees (GBDT), the second-best model identified in Table 8, and the empirical analysis is re-executed using Shapley Additive Explanations. Second, the study employs an alternative interpretable machine-learning approach—the Explainable Boosting Machine (EBM)—to replicate the empirical results. EBM is particularly suited for capturing nonlinear relationships (Nazemi et al., 2022; Nazemi and Fabozzi, 2024). Third, normalization is incorporated into the objective function to reduce model complexity, mitigate overfitting and accelerate the learning process (Jabeur et al., 2024). All robustness tests are consistent with the findings of this paper, underscoring the reliability and robustness of the study.

Given the limited research on this topic, this paper makes significant contributions to the literature on the monetary base distribution of collateral monetary policy, the shadow banking system and bank risks. Understanding bank risks within the context of the existing financial system requires an examination of money creation (Jakab and Kumhof, 2018). Our study highlights that comprehending bank risks begins with understanding unregulated money creation. Informal financial activities that allow banks to circumvent financial regulations and create money can potentially lead to financial instability, with the shadow banking system serving as a prominent illustration. Furthermore, this paper enriches the literature on collateral monetary policy. The findings suggest that disparities in liquidity provision across different banks lead to insufficient liquidity, prompting them to engage in informal financial activities. The design and implementation of collateral monetary policy instruments must strike a balance between the liquidity demands of various banks. Failing to do so creates artificial liquidity gaps for non-primary banks, which in turn incentivizes unregulated banking behavior.

The paper also contributes to a better understanding of China’s 2018 NAM Regulations and their implications for the design of financial regulatory frameworks. Given the money-creation function of WMPs, as discussed in Section 2 (Sun, 2019), these products should be regulated as loans under the macro-prudential framework. Thus, the study of the relationship among collateral monetary policy, the shadow banking system and bank money creation links them with macro-prudential financial regulations—an area not adequately addressed in the existing literature.

Overall, the findings emphasize the biased effect of collateral monetary policy on liquidity distribution, which facilitates the growth of the shadow banking system and unexpectedly increases bank risks. This paper enhances our understanding of the underlying drivers of the shadow banking system (Chen et al., 2018) and the real economic consequences of collateral monetary policy (Nyborg, 2017; Yuan et al., 2024). Additionally, the exploration of collateral monetary policy contributes to financial risk management and the design of optimal monetary policy instrument combinations for central banks (Kim and Chen, 2022).

This paper highlights the far-reaching implications of collateral monetary policy on the shadow banking system and the bank risks exacerbated by its inherent liquidity distribution bias. By examining the case of China, this paper reveals that the shadow banking system flourishes as a mechanism to mitigate the adverse effects of the liquidity allocation bias inherent in collateral monetary policy, thereby producing unanticipated consequences for financial stability. These outcomes, in turn, contribute to the negative repercussions of collateral monetary policy. These insights contribute to the literature on the design of monetary policy instruments in promoting financial stability (Cesa-Bianchi and Rebucci, 2017; Bussière et al., 2021). Moreover, the NAM Regulations in China effectively curtail the money creation activities by the shadow banking system, thereby mitigating associated bank risks.

This section presents the results from SHAP-GBDT to show the impact of the collateral monetary policy on bank risks.

Figure A1 

Figure A2 

This section presents the results from EBM to show the impact of the collateral monetary policy on bank risks.

Figure A3 

This section presents the results with normalization data to show the impact of the collateral monetary policy on bank risks.

Figure A4 

Figure A5