Shadow Banking and Regulatory Arbitrage

1

Shadow Banking and Regulatory Arbitrage: Evidence from China

Xiaoli Wan¹ University of Auckland

Abstract: This paper examines the fundamental factors of the evolution of China’s shadow banking based on some unique institutional features of China’s money and banking system.

By manually collecting the issuance amount data of wealth management products(WMPs), the main funding source of China’s shadow banking, we find that the macro policy arbitrage, including risky loan regulation and interest rate constrain are the most important for banks issuing WMPs rather the micro prudential regulatory policies, such as capital adequacy requirements or loan-to-deposit ceiling. Tightened credit quota control can actually constrain banks’ WMPs with a stronger effect on big 4 state banks. Unexpectedly loosening regulation in loan scale, risky industry can largely reduce WMPs. Small banks usually shrank one third more than big 4 banks, indicating that small banks are more independent on loan business, especially the loans in the risky industries.

1 Business school, University of Auckland. Email: [email protected]. Thanks to Dimitris Margaritis and Helen Lu.

2

I. Introduction

Since the outbreak of the global financial crisis (GFC) in 2007, an increasing strand of literature explores how shadow-banking sectors arise and expand rapidly. Regulatory arbitrage is commonly viewed as the key reason. Much of this literature focuses on the regulatory capital arbitrage (RCA) of shadow-banking system in developed countries, while little research paid attention to the developing markets. As the second largest economy in the world, China’s shadow banking system did expand massively since 2009 when the People's Bank of China (PBC) tightened monetary policy after China's government implemented the large-scale stimulus program in 2008 to offset the negative shocks of GFC.

Although there is no exact and standard definition on China’s shadow banking yet, the general consensus is that the credit caused by various off-balance operations of traditional banks is the key part. Different with the massive collateralisation or securitisation in the U.S.

shadow banking system, China's is better characterised as a bank-like credit intermediation process (Torsten Ehlers, 2018). Figure 1 shows the evolution of total bank credit via the RMB loans and off-balance activities. It is obvious that the ratio of bank loans as of total bank credit has exhibited an obviously descending trend, from 86% in 2006 to 64% in 2013, whereas this ratio rebounded to 95% in 2015 and fell to 80% in 2017. The evolution of this ratio can be partly justified from the change of on-balance and off-balance liquidities. Figure 2 displays the year-on-year growth of monthly new lending. During the year between 2009 and 2011, the change of loans are negatively correlated with the change of off-balance liquidity, while they were positively correlated during some periods.

Figure 1. Total bank credit and RMB loans

60 65 70 75 80 85 90 95 100

0 5 10 15 20 25 30 35 40

2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 RMB Loans

Total Bank Credit

Loans/Total Banks Credit(RHS)

Ratio(%GDP） Ratio（%）

3

Figure 2. Growth of traditional bank loans and off-balance credit

What arose the rapid development of China’s shadow banking system and what is the main driven factors of the evolution of the off-balance credit? This paper aims at answering these questions from the perspective of regulatory arbitrages by examining the major component of China's shadow banking sector--wealth management products (WMPs) issued by banks.

Compared with the previous work focusing on the Regulatory Capital Arbitrage (RCA) in developed countries, we pay more attention to the unique regime arbitrage channels in China, especially the macroeconomic policy arbitrage including monetary policy and credit policy arbitrage.

Compared to the other central banks in the world, China's central bank, i.e., the PBC, has executed both monetary policy and credit policy. For the monetary policy, the PBC announced that its manipulation has been transferring from the quantity-based to price- based. Since 1996, the PBC has began to use open market operations to form the market- based interest rates, but still administratively decide the policy rates including benchmark interest rate and lending rate at the same time, leading to a dual-track interest rate system in China. Meanwhile, the PBC still views the M2 growth as its intermediate target of monetary policy. At the same time, the PBC has separate credit policy, mainly including credit quota control and credit allocation guidance. For example, the PBC restricted banks lending loans into real estate industries when the central government aimed at containing housing price, or to local government platform when concerned about the debt burden of local governments.

Although these policies are macro policy, they have similar impacts on banks with regulatory policies as most of the policies are administratively executed. So commercial banks have incentives to do the regulatory arbitrage by various shadow banking activities.

However, these policies might have different effectiveness on state banks and non-state banks since the state banks are directly controlled by the central government.

Recently, there are a few studies on China's shadow banking system. Dang, Wang, and Yao (2014) provides a theoretical model to explain the differences between China and the U.S. shadow banking activities. Hachem and Song (2015) build a model to show that the asymmetric competition between big and small banks is the key for China's shadow banking development. Allen, Qian, Tu, and Yu (2019) and Chen, Ren, and Zha (2018) study another component of China's shadow banking, i.e., entrusted loans. Acharya, Qian, and Yang (2017) empirically show that the major component of China's shadow banking sector—wealth management products(WMPs) seems to be triggered by the stimulus plan launched by

-5000 -4000 -3000 -2000 -1000 0 1000 2000 3000 4000 5000 6000

2006S1 2008S1 2010S1 2012S1 2014S1 2016S1 2018S1 On-balance liquidity Of f -balance liquidity New Credit Growth（Billion RMB, yoy)）

4

China’s government in the end of 2008. Overall, these literatures focus on impacts of deposit competition or deposit withdraw risks, either caused by the tightened monetary policy or the unexpected stimulus package of China's government. Furthermore, most of these studies only explained why China's shadow banking sectors expand so fast while few paid attention to the volatile growth of these off-balance liquidities.

To understand the fundamental driven factors of China's shadow banking, we use semi- annual data on the main listed commercial banks of China between 2008 to 2017. Despite the various intermediate channels innovated by China's shadow banking system including bank-trust cooperation, bank-security brokerage cooperation, bank-mutual fund cooperation, etc., the shadow banking funding mainly relies on the WMPs. By manually collecting the issuance amount of WMPs of these banks rather the number of WMPs issued which has been used in most of the previous literature, we empirically explore the driven factors of banks issuing WMPs. First, we find that the macro policy arbitrage, including risky loan regulation and interest rate constrain are the most important for banks issuing WMPs rather the micro prudential regulatory policies, such as capital adequacy requirements or loan-to- deposit ceiling². Second, tightened shocks of credit quota control can actually constrain banks’ WMPs with a stronger effect on big 4 banks. More stringent regulation on risky industry loans and higher spread between market and policy interest rates would stimulate WMPs and these impacts are not significantly different for big 4 (state-banks) and the other small banks. Thirdly, unexpectedly loosening regulation in loan scale, risky industry can largely reduce WMPs. Small banks usually shrank one third more than big 4 banks, indicating that small banks are more independent on loan business, especially the loans in the risky industries.

Our paper contributes to and extends the literatures on the formation of shadow banking and interaction between shadow banking sectors and regulations from a longer and dynamic aspect. First, we verify that the macroeconomic policy arbitrage of China's banks is the fundamental factor other the stimulus plan in 2009. WMPs as the major shadow-banking component with the longest history in China, was initially available at the end of 1990's, but have not developed rapidly until the year of 2005 when China's government launched the first round of large-scale manipulation constraining domestic housing price. From 2005 to 2008, the market size of WMPs has already increased sharply from 356 billion RMB to 926 billion RMB. Second, we explain the volatile evolution other than the expansion of China's shadow banking system in the last decade. Because of the change of macroeconomic policy stance, the development of China's shadow banking system also exhibits a dynamic pattern.

Thirdly, we explore the micro channel of macro policy arbitrage and identify the heterogeneity of the responses of banks by using bank-level dataset. Finally, we are the first to explore the asymmetric impacts of macro policy on shadow-banking sectors.

The rest of the paper is organized as follows: in section 2, we provide some institutional backgrounds about China's shadow banking and the framework of China's monetary policy and credit policy. We then outline the hypotheses for subsequent empirical testing in section

2 China's banks had been required to keep their loan-to-deposit ratio lower than 75% for twenty years, which was cancelled in 1st, October, 2015. However, this ratio is still the monitoring index of the PBC.

5

3. In section 4, we describe the data and variable definitions and summary statistics. Section 5 reports the econometric methods and discuss the results. Finally section 6 concludes.

II. China’s Shadow Banking and Manipulation of Macroeconomic Policy

In this section we provide a brief introduction about China's institutional background on several unique features of China's monetary framework, banking system and regulation. We focus on three issues: (a) how the PBC executes its policies; (b) evolution of China's shadow banking system; (c) characteristics of state and nonstate banks in China.

A. Monetary Policy and Credit Policy

Different with the other central banks in advanced economies, China’s central bank, the PBC, not only execute traditional monetary policy (quantity-based monetary policy: M2 as the intermediate target), but also has separate credit policy including credit quota control and credit allocation guidance. For example, the PBC restricted bank loans to real estate industries when the central government aimed at containing housing price, or to local government platform when concerning about local government debt burden.

The unique characteristic comes from the development history of the PBC. As is well known, China's economy was centralized. The so-called central bank has not been established until the middle of 1990's. The PBC directly controlled bank loans and credit allocations before 1993. Since 1996, it began to transfer a standard central bank with money supply as an instrument to operate monetary policy, but still directly controlled the bank loans. Till 1998, the PBC cancelled the direct control on bank loans and officially announced that M2 growth was its intermediate target. Open market operation and required reserve ratio were the two main tools for PBC to complement its quantity based monetary policy.

However, the PBC still keep its implicit impacts on bank loans through so-called window guidance. Almost after the spring festival of every year, the PBC would issue a credit quota for every main banks to control the credit scale, especially for the big 5 state banks.

Meanwhile, the PBC manage the credit quality by regulating credit allocation of banks.

During the last decade, the PBC mainly concerned about the risk from the real estate bubbles and some overcapacity industries since the early of 2000's. After the stimulus package implemented by the government in 2009 to offset the negative shock of the global financial crisis, the debt burden of local governments increased sharply through all kinds of platforms owned by local governments. However, banks have high incentive to allocate loans to these industries, either because of the relatively high return and low risk from the industries themselves or the implicit issuance from the local governments. So the loans to real estate industry, overcapacity industries and local government platform have been regulated by the PBC intensively. Following Chen et al., (2018), we call this quality-control of banks loans as safe-loan regulation (or risky-loan regulation). It is noted that the intensity of this safe-loan regulation has been adjusted according to China's macro economy momentum. When the economy growth was uncertain or too weak, the PBC might ease the control on these risky loans. For example, the loans to real estate industry have not been restricted from the end of 2008 to the end of 2009, leading to a distinguished increase of real estate loans shown in Figure 3.

6

Figure 3 RMB loan growth and bank loan/all external funding of real estate industries Another important characteristic of China’s monetary framework is the dual-track interest rate system. Although PBC has launched a series reform to make interest rate more market- oriented, however, the dual-track interest rate system still exists practically. PBC had set base interest rates along with upper and lower bounds for a long time. Nevertheless, the lower bound of lending rate and the upper bound of deposit rate have been cancelled in August 2013 and October 2015 respectively. Since the end of 2015, PBC only decide the base interest rate, while the interest rate alliance among banks kept the base rate playing a key role practically.

B. Evolution of China’s Shadow Banking

The forms of shadow banking in U.S. and China differ significantly. China’s shadow banking activities entail direct or indirect lending and are often linked to banks, while shadow banking in the US is dominated by complex derivatives such as securitized loans, asset-backed commercial paper, repurchase agreements, and money market funds (C. Li, 2013). Essentially, China's shadow banking is the "shadow of banks"(Torsen et al., 2018).

Banks expand their credit through off-balance activities in which securitization and market- based instruments play only very limited role. As can be seen from Figure 4, the growth rate of on-balance bank loans tracked with M2 growth rate closely before 2005. Since then, the pattern of these two indicators began to diverge, which has been viewed as part of the evidence of the increasing importance from shadow banking credit.

-100 -50 0 50 100 150 200 250

-2 -1 0 1 2 3 4 5

06 07 08 09 10 11 12 13 14 15 16 17 18

Dev from Trend for Loan/All Funding in Real Estate Industry (RHS) Dev from Trend for RMB Loan Growth

7

Figure 4. Growth rate of on-balance loans and M2

Banks have shifted their credit channels at the intermediate stage a lot, including so-called bank-trust cooperation, bank-security brokerage cooperation, bank-mutual fund cooperation, etc. However, the funding of China's shadow banking rely all along on depositors, who purchase bank-issued WMPs or trust products. While the WMPs played much more important role than trust products in China's shadow banking and are connected with the majority of the depositors, we focus on the evolution of WMPs in China.

The development of WMPs has experienced three stages in China. The first is the embryonic stage before 2005. The first WMP was issued in 2002. Since 2004, each bank has launched its own financial products one after another. At this time, the product structure was relatively simple. Basically, banks transferred parts of their low-risk investment income to customers under some conditions of investment threshold and liquidity. The second is the fast growth stage between November 2005 to the middle of 2008. The main characteristics are the increasing number of WMPs issued and the increasing variety of product types. The funding scale by WMPs increased from 356 billion RMB in 2005 to 926 billion RMB in the first half of 2008.The third stage is from the middle of 2008 to the present. During this period, massive regulations about WMP market have been released. The growth momentum of the whole WMPs market slowed down since the end of 2016 after a sharply increase. Till the end of 2017, the capital raised by WMPs were 174 trillion RMB, only 10.1% increase compared to 2016.

C. State versus Nonstate Commercial Banks (Big 4 versus Small Banks)

The main micro-prudential regulations for China's commercial banks include capital adequacy requirements (CAR) and loan-to-deposit ceiling, i.e., the ratio of bank loans to bank deposits should be less than 75%. LDR regulation was established in 1994 to manage the quantity of bank loans. Anyway, with the background of the national international finance changing, the overseas liquidity for the whole banking system is descending.

Meanwhile, the competition between banks become intensive. Some small banks face more challenges to get the enough deposit. The LDR regulation was cancelled by the PBC at

5 10 15 20 25 30 35

02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 Loan Growth (yoy) M2 Growth (yoy)

8

October, 2015. So we separate the whole sample period (2008-2017) into two subgroups, before and after October, 2015. Table 1 shows the summary statistics of the main regulatory and financial indicators.

Table 1. Summary Statistics of Bank Characteristics

All Sample N Mean Stdev Min Max

Big 4 Banks

WMPs Issuance/Asset 80 16.659 7.806 .686 40.549

CAR 80 13.082 1.449 8.31 15.58

Ldratio 80 65.173 6.844 49.5 79.782

ROA 80 .601 .119 .306 .86

NIM 80 2.474 .331 1.83 3.29

Small Banks

WMPs Issuance/Asset 160 34.29 32.48 .373 176.155

CAR 160 12.176 2.33 8.11 25.59

Ldratio 160 70.237 9.877 42.84 105.157

ROA 160 .55 .144 -.334 .979

NIM 160 2.525 .427 1.4 3.66

Before 2015Oct N Mean Stdev Min Max

Big 4 Banks

WMPs Issuance/Asset 60 16.797 8.395 .686 40.549

CAR 60 12.663 1.347 8.31 14.8

Ldratio 60 63.801 5.952 50.84 74.16

ROA 60 .623 .119 .306 .86

NIM 60 2.565 .309 2.04 3.29

Small Banks

WMPs Issuance/Asset 120 27.257 27 .373 176.155

CAR 120 12.096 2.619 8.11 25.59

Ldratio 120 68.68 7.141 42.84 84.063

ROA 120 .581 .148 -.334 .979

NIM 120 2.649 .372 1.85 3.66

After 2015Oct N Mean Stdev Min Max

Big 4 Banks

WMPs Issuance/Asset 20 16.245 5.852 5.853 35.697

CAR 20 14.341 .935 12.33 15.58

Ldratio 20 69.291 7.808 49.5 79.782

ROA 20 .534 .091 .395 .703

NIM 20 2.2 .23 1.83 2.66

Small Banks

WMPs Issuance/Asset 40 55.389 38.311 11.548 148.311

CAR 40 12.415 1.059 10.8 15.48

Ldratio 40 74.906 14.571 46.76 105.157

ROA 40 .455 .077 .333 .65

NIM 40 2.152 .366 1.4 2.79

Both state banks and small banks met the capital requirements by a comfortable margin as shown in table 1. Some small banks even have much higher CAR than state banks. LDR is getting higher after October, 2015 for both big 4(state) and small (nonstate) banks. Before October 2015, the mean of LDR is less than the 75% ceiling not only for state but also for nonstate banks. Anyway, some small banks sometimes did exceed the ceiling. Nevertheless, the difference of ROA, NIM between big 4 banks and the others are not significant. In

9

summary, both state and nonstate banks meet the major policy requirements during the sample well.

It is interesting that WMPs issuance, measured as the ratio of total asset, is decreasing slightly from 16.8% to 16.2% after 2015 for big 4 banks, while increasing significantly from 27% to 55% for small banks. Since the regulatory requirements do not seem to explain the difference pattern between state and nonstate banks, then would it be explained by the institutional features? The big 4 banks are directly controlled by central banks and the other banks are not. We will identify the impact of the institutional difference on banks shadow banking activities in the empirical section.

III. Theory and Hypothesis

We classified the possible main driven factors of China’s shadow banking sectors into two categories: micro- and macro- regulatory arbitrage. The former one is the regulation complemented by China’s Banking Regulatory Commission (CBRC), and the latter is mainly executed by the PBC, related with the macroeconomic policy manipulation.

A. Regulatory Capital Arbitrage (RCA)

RCA is the main arbitrage mechanism discussed in the literature about shadow banking in developed countries. As Capital Adequacy Ratio (CAR) is the key regulatory indicator, banks can improve their CAR without reducing the risk of lending activities through complex derivatives such as securitized loans etc. However, China’s banks make RCA by off-balance lending through trusts, insurance or wealth management companies which are not restricted by CAR. But in our sample, both state banks and small banks met the capital requirement by a comfortable margin as shown in table 1. So we have the first hypothesis as following:

Hypothesis 1:

RCA is not important for China’s shadow banking as China’s banks go far beyond the capital requirement.

B. Loan-Deposit-Ratio (LDR) Arbitrage

Differently with the most common banking regulations in other countries, there is an additionally regulatory policy in China, i.e. the loan-to-deposit ratio of banks cannot exceed 75%. The initial aim of this measure is to restrict the excess risk-taking behaviours of the bank lending and control the systemic risk as China’s financial system is dominated by banks. Most of banks did not view this ratio as a restriction since there had been huge savings in China’s banking system. However, this ratio might become binding for some small banks with more intensive competition in bank industry.

From the perspective of the national banking system, the banks have the liquidities to lend only if they can constantly absorb and create savings. Before the GFC, China accumulated a huge amount of foreign reserves by its export-led economy. During the year of 2002 to 2007, one of the main jobs of PBC, China’s central bank, is to withdraw the passive RMB injection from the market by the quick accumulation of foreign reserves.

However, the export growth decreased sharply since GFC and China’s economy has transformed to be invest-led then. The foreign reserve began to decrease in recent years.

10

Meanwhile, with the RMB exchange rate more flexible, loan-to-deposit ratio became binding for China’s banking system, especially for those small banks. This change might lead to the banks to issue large amount of wealth management products so as to meet the LDR criteria either from transferring some on-balance lending to off-balance lending or expanding deposit by internal transfer funds from WMPs. China's government has already realized the problem of LDR regulation, so the LDR criteria is weakened to a liquidity monitoring indicator rather the regulatory indicator from October of 2015.

Hypothesis 2a:

When LDR was a regulatory indicator before October 2015, banks with LDR reaching or exceeding the 75% ceiling would operate more off-balance activities as LDR was higher.

Hypothesis 2b:

Banks with LDR at a safe level would issue less WMPs when LDR is higher, since they have enough room to expand business by on-balance lending rather off-balance lending.

C. Macroeconomic Policy Arbitrage: Monetary Policy and Credit Policy

Strictly speaking, the macroeconomic policy cannot be viewed as regulations. However, as most of the manipulations are administrative in China, the impacts of macro policies on banks' activities can be regarded as regulatory arbitrage.

To date there are few literatures which discussed about the impact of monetary policy on banks' off-balance credit. As Hakor(2005) and Morgan(1998) pointed out, tightened monetary stance could stimulate the off-balance activities of banks as it became more difficult to access the bank loans. On the other hand, tightened monetary policy would also result a higher external financing cost, or reduce the liquidity in bank system (such as a hike in required reserve ratio), so as to constrain the off-balance liquidity either. Credit quota played a key role for PBC to target its monetary aggregate goal. A smaller credit quota usually means a tightened monetary stance.

Hypothesis 3a.

A tightened monetary policy stance would restrain the shadow banking activities by constraining the total liquidity condition.

Hypothesis 3b.

A tightened monetary policy stance would stimulate the shadow banking activities by constraining the on-balance lending quota.

China’s central bank has a tradition to impact or control the credit allocation of commercial banks, which is called as macro prudential credit policy or part of the industry policies. The RMB loans quotas³ and the restrictions of loans in "risky" industries are the two key credit policies to impact banks’ lending behaviour, according to a report of ICBC, the biggest commercial bank in China. This pattern is extremely obvious since 2010. The large scale stimulating packages complemented by China's government since the GFC was mainly to expand investments in infrastructures and real estate industry, leading to the

3 The quota for banks means the maximum amount of loans a bank can issue in one year.

11

sharply increase of local government debts and housing bubbles since 2008. In order to control the financial systemic risks, PBC began to tighten the monetary stance and restrict the banks to provide loans to local government platforms, real estate companies and other over-capacity industries. However, given the implicit insurance of central government for local government debts and the high return of housing industry or other over-capacity industries, banks would provide off-balance liquidities to these industries when they face the pressure from lower profitability on on-balance assets.

Hypothesis 4.

If there are more stringent restrictions on loans in "risky" industries, there will be more off-balance credit when banks have a lower profitability on on-balance assets.

Although China’s monetary policy is still essentially quantity based(Chen et al. (2018), PBC has endeavoured to transfer to be price based. The dual-track interest rate system actually reflects the process. With China’s banks facing a more competitive environment, the interest rate regulation gives banks strong incentive to engage shadow banking business (Acharya et al., 2017). As the market rate has been almost higher than the base interest rate, banks want to attract more funding by issuing WMPs which are not regulated by the base interest rate. As shown in figure 5, Spread3M is defined as the difference between market interest rate (Shibor with 3 month maturity) and base interest rate(3-month deposit rate).

Higher spread indicates a severe extent of interest rate constrain. So we expect higher spread would stimulate WMPs issuance of banks.

Figure 5. Base interest rate and market interest rate

Hypothesis 5.

Stronger interest rate restrains (higher spread) would stimulate shadow banking activities.

-1 0 1 2 3 4 5 6 7

02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 DR3M

SHIBOR3M SPREAD3M

12

IV. Dataset and variable definitions

The dataset we manually constructed is a semi-annual panel dataset of WMPs issuance of publicly listed banks. The WMPs issuance information is most important for this paper.

Previous literature about WMPs mainly use the number of WMPs issued by banks, rather the issuance amount of WMPs. There are 19 banks listed in Hong Kong, Shenzhen or Shanghai Exchange. However, as the WMPs data is not a compulsory information to be released by banks, only 12 of them have relatively complete information about WMPs issuance. Even for one of the five big state banks in China, Jiaotong Bank did not release its WMPs issuance data before 2013. The 12 banks we collected include big 4 state banks; the rest are all nonstate banks. Their assets take up, on average for 2008-2017, 54.27% percent of total bank assets and 43% percent of total WMPs issuance in the entire banking system.

These banks can be a satisfied representative of china’s banking system based on the data availability.

We read through annual reports and collected the data about WMPs issuance from the financial reports of listed banks. Although the WMPs has increased rapidly since 2005, most of the data about WMPs issuance at the bank level is only available from 2008 when the government began to concern the risks of shadow banking sectors. So our dataset covers from 2008 to 2017 for the 12 listed banks. Standard bank indicators, including assets, CAR, LDR, ROA etc. are collected from Wind database. Other macro data is from CEIC database.

For the variables to represent the policy effect, it is necessary to extract exogenous part form the endogenous part as in the empirical macroeconomic literature ((Leeper et al. 1996;

Christiano, Eichenbaum, and Evans 1999, 2005; Sims and Zha 2006). In order to compare our results with the previous literature, we estimate equation (1) in below following Chen, etc. (2018). g_m,t is M2 growth rate as the proxy of the aggregate monetary policy stance. We then use the residual (ExM2) of the equation to represent China's monetary policy shock only considering the inflation gap and output gap.

g_m,t=γ₀+γ₁g_m,t-1+γ₂(π_t-1−π^*) +γ₃(g_𝑥,t-1− 𝑔_{𝑥,𝑡−1}^∗ ) + 𝜀_𝑚,𝑡 (1) Different with Federal Reserve or European central bank, China's central bank has credit policy including credit scale control and risky loan regulation. Specially, credit quota of the banking system is adjusted by the central bank according to the fundamentals, including inflation and GDP growth. So we estimate a similar credit scale control policy rule by using on-balance lending growth rate as the proxy and use the residual (Exloan) to represent the credit scale policy shock.

g_c,t=γ₀+γ₁g_𝑐,t-1+γ₂(π_t-1− π^*) +γ₃(g_𝑥,t-1− 𝑔_{𝑥,𝑡−1}^∗ ) + 𝜀_c,t (2) Risky loan regulation, as another credit policy tool, is mainly used by the central bank to constrain the credit allocation into the risky industries, such as the real estate industry and local government platforms. As the data about local government platforms is hardly to be obtained, we create a proxy to capture the risky loan regulation in real estate industry. For real estate developers in China, they usually prefer bank loans to the other external funding sources as the cost of bank on-balance loans is the lowest. So, we use the deviation from the trend of “bank on-balance loans/total external funding” in real estate industry to measure the

"risky loan" regulation policy. The lower the deviation (Reloans) is, stricter regulation aimed at risky loans is.

13

Dual-track interest rate is another important feature of China's monetary policy. Central bank has administratively announced the benchmark deposit and lending rate and use the open market operation to impact the interbank interest rate as the Federal Reserve does.

During most of the time, the market-oriented interest rate, i.e., the interbank interest rate named as Shibor, is higher than the administrative interest rate, i.e., the benchmark deposit rate. Following Acharya etc. (2017) we define Spread as the gap between market interest rate (3-month Shibor) and base policy rate (3-month deposit rate) to capture the extent of interest rate restrain. Higher spread indicates a more severe interest rate constrain.

Since M2 growth reflects the ultimate results after the PBC manipulates both the standard monetary policy tools (open market operations and required reserve ratio etc), the credit policy and the interest rate constrain, the residual of equation (1), i.e., ExM2 actually contain the shocks of credit policy and interest rate constrain. So, we estimated an extended version of equation (1) by considering the other shocks of China's central bank including credit scale shock, risky loan regulation shock and interest rate constrain shock. The residual (ExM2ext) of equation (3) is used by us to capture the monetary policy shock excluding the impact of China's credit policy and dual-track interest rate regime.

g_m,t=γ₀+γ₁g_m,t-1+γ₂(π_t-1−π^*)+γ₃(g_𝑥,t-1− 𝑔_{𝑥,𝑡−1}^∗ )

+𝐸𝑥𝑙𝑜𝑎𝑛_𝑡+ 𝑅𝑒𝑙𝑜𝑎𝑛𝑠_𝑡+ 𝑆𝑝𝑟𝑒𝑎𝑑_𝑡+ 𝜇_𝑚,𝑡 (3) Table 2 summarizes the data sources and variable definitions for the key variables.

Table 2. Variables Definition and Data Source

Variable Definition Source

Shadow WMPs Issuance scaled by Bank Asset Annual Reports, Interim Reports

CAR Capital Adequacy Ratio Wind

Ldratio Loans/Deposits Wind

Exm2 The residual of equation (1) and normalized by its standard deviation.

CEIC;

Author’s calculation Exloan The residual of equation (2) and normalized by its standard

deviation.

CEIC;

Authors’

calculation Reloans For real estate developers in China, they usually prefer

bank loans to the other external funding sources as the cost of bank loans is the lowest. So we use deviation from the trend of “loans/total external funding” in real estate industry to measure the "risky loan" regulation policy. The lower Reloans is, stricter restrictions hold for “risky loan”.

CEIC;

Authors’

calculation

14

Normalized by its standard deviation.

Spread Shibor3m-deposit rate 3m. Normalized by its standard deviation.

CEIC

ExM2ext The residual of equation (3) and normalized by its standard deviation.

Control Variables

ROE Return of Equity. Zhu et al. (2016) suggested that ROE measures the whole profitability of banks for shareholders.

Banks with lower ROE have higher incentive to develop shadow banking business.

Wind

NIM Net interest income/Interest earning assets, Pan and Wei(2015), Guo and Zhao(2015) suggest that lower NIM means a weaker profitability on loans for banks, which led to a higher incentive for shadow banking activities.

However, we think it might not be true in China. If banks do have a high NIM but faced a loan quota restriction or risk loan regulation, higher NIM may induce banks to increase off-balance lending.

Wind

NIIR Non-Interest income/total operating income. NIIR indicates the profitability of banks through non-interest income or the dependency of traditional deposit and loan business of banks. Banks with higher NIIR might have more incentive to operate off-balance activities.

However, when banks face tightened macro policy including tightened credit quota, strict risky loan controls, higher spread, banks which was more dependent on traditional loan business, i.e., lower NIIR, would have higher incentive to operate shadow banking activities to earn from off-balance lending.

CEIC;

PBC’s Website;

Authors’

Calculation

GDP year over year real GDP growth rate CEIC

Infl year over year inflation CEIC

15

V. Empirical results and analysis

1. Which is more important? Micro-prudential or Macro-prudential regulations?

From the micro regulation aspect, China’s banks face the capital adequacy requirements and LDR regulation. From the macro level, China’s banks are impacted by economic momentum and related macroeconomic policies, including the tightened or loosened monetary policy, “risky loan” regulation imposed sometimes. Whether the micro regulation policy or macro policy arbitrage is more important for driving China’s shadow banking sectors? To answer this question, we firstly estimate the equation (4) which only includes the micro regulation policies, then we estimate equation (5) which includes the macro regulation policies⁴.

𝑆ℎ𝑎𝑑𝑜𝑤_𝑖𝑡 = 𝛼 + 𝛽₁𝐶𝐴𝑅_𝑖𝑡−1+ 𝛽₂𝐿𝑑𝑟𝑎𝑡𝑖𝑜_𝑖𝑡−1+ 𝛽₃𝐻𝐿𝐷𝑅_{𝑖,𝑡−1}+

𝛽₄𝐻𝐿𝐷𝑅_{𝑖,𝑡−1}× 𝐿𝑑𝑟𝑎𝑡𝑖𝑜_𝑖𝑡−1+ 𝑏𝑎𝑛𝑘_𝑖+ 𝑇𝑖𝑚𝑒_𝑡+ 𝑒_𝑖𝑡 (4) Where i refers to bank and t refers to the end of every half year. HLDRt is a dummy variable indicating the LDR regulation is binding for banks, which equals to 1 if Ldratio is higher than 72.33%, otherwise 0. We chose the threshold as 75% percentile in the sample period when LDR was a regulatory indicator before October 2015.

In order to explore the effect of macro policies arbitrage, we estimate the following regression:

𝑆ℎ𝑎𝑑𝑜𝑤_𝑖𝑡 = 𝑐 + 𝑐₁𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡+𝛽₁𝐶𝐴𝑅_{𝑖,𝑡−1}+ 𝛽₂𝐿𝑑𝑟𝑎𝑡𝑖𝑜_𝑖𝑡−1+ 𝛽₃𝐻𝐿𝐷𝑅_{𝑖,𝑡−1}+

𝛽₄𝐻𝐿𝐷𝑅_{𝑖,𝑡−1}× 𝐿𝑑𝑟𝑎𝑡𝑖𝑜_𝑖𝑡−1+ ∑²_𝑙=1𝜑_𝑙𝑀𝑎𝑐𝑟𝑜_𝑐𝑜𝑛𝑡𝑟𝑜𝑙_𝑙,𝑡+ 𝑏𝑎𝑛𝑘_𝑖+ 𝜀_𝑖𝑡 (5) Where Macro control variables includes GDP growth rate and inflation rate.

Table 3. Estimation results for Micro- v.s. Macro- regulation arbitrage

(1) (2) (3)

L.CAR 0.009 -0.003 -0.043

(0.017) (0.012) (0.027)

L.Ldratio -0.011^* -0.005 -0.018^**

(0.006) (0.005) (0.009)

Hldr -2.295^* -0.253 1.831

(1.328) (0.897) (2.795)

Hldr×L.Ldratio 0.032^* 0.004 -0.025

(0.018) (0.012) (0.037)

4 Monetary policy is usually a macro policy rather than a regulatory policy. However, China’s monetary policy is not purely market towards. For example, china’s central bank(PBC) can control the total bank lending quota by so-called “window guidance”. Additionally, PBC can also control the credit allocation through “risky loan regulation”. The real estate loan and loans to local

government platform were the main risky loan when risky loan regulation was tight. What’s more, China’s interest rate system is still dual-track. PBC can decide the benchmark deposit and lending rate directly. So China’s monetary policy can be viewed as macro regulatory policy.

16

ExM2 -0.273^***

(0.032)

Exloan -0.242^***

(0.067)

Spread 0.291^**

(0.116)

Reloans -0.431^**

(0.203)

ExM2ext 0.195^***

(0.061)

Reloans×L.ROE 0.0249^***

(0.008)

L.ROE -0.0356^***

(0.013)

GDP -0.118^*** -0.184^**

(0.021) (0.077)

Inflation -0.001 0.055

(0.018) (0.080)

C 2.302^*** 4.251^*** 6.341^***

(0.462) (0.469) (1.063)

Bank fixed effects Yes Yes Yes

Time fixed effects Yes No No

Observations 228 228 228

N 228 228 228

Standard errors in parentheses

*p< 0.1, **p< 0.05, ***p< 0.01

Column (1) of Table 3 shows the results of equation (4) based on the two-way fixed effects estimation. The coefficient of CAR is positive but insignificant. The coefficient of Ldratio is negative, which indicates that banks with higher Loan-to-deposit ratio (Ldratio) in the last period actually issue less WMPs if the Ldratio regulation is not going to be binding or Ldratio is not a regulatory indicator. The coefficient of the interaction term between HLDR and Ldratio is positive at the 10% significant level, which indicates that banks with higher LDR would issue more WMPs when their LDR is reaching the 75% ceiling. The total impact of HLDR is significantly positive when Ldratio is higher than 71.7%, which is satisfied if HLDR is not zero. These results shows that WMPs is more likely driven by Ldratio regulatory when banks’ Ldratio is going to reaching the 75% ceiling before October, 2015, while the capital adequacy arbitrage is not important for China’s shadow banking.

17

Column (2) to (3) show the results when the macro policy is considered. In these two cases, the coefficients of CAR and HLDR become insignificant, and the coefficients of macro policy are significant. As the theory predicted, the macroeconomic policies arbitrage exists significantly and is more important than the micro-prudential regulations such as CAR and LDR requirements. In column (2), we use ExM2 to proxy the monetary policy shock which contains all the other shocks including credit policy and interest rate constrain and the others. A negative shock would significantly stimulate WMPs of banks.

In order to identify the impacts of the detailed shocks, we use the credit scale control shock (ExLoan), risky loan regulation shock(Reloans), interest rate regulation shock (Spread) and the net M2 growth shock (ExM2ext) rather than the aggregate monetary policy shock(ExM2) to proxy the macro policy in equation (5). The results are shown in column (3). As the hypothesis expected, tightened credit quota and risky loan regulation shock, or higher spread would stimulate more WMPs. The coefficient of the net monetary policy shock (ExM2ext) is significantly positive. This suggests that tighten monetary policy shock did restrain the WMPs, but just existed after excluding the impacts of the credit controls and interest rate regulation.

In summary, these results generally support that China’s shadow banking is essentially driven by macroeconomic policies arbitrage, rather than the micro regulation arbitrage.

2. The heterogeneous impact of macro policy on WMPs:

In order to examine whether the impact of macro policy on shadow banking would be different for the state and nonstate banks, we define a dummy variable(MS) which equals to 1 if the bank is not one of the big 4 state banks, otherwise 0. Then we estimate the following equation (6) by including MS and the interaction term of macro policy and MS.

𝑆ℎ𝑎𝑑𝑜𝑤_𝑖𝑡 = 𝑐 + 𝑐₁𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡+ 𝑐₂𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡× MS + 𝑐₃𝑀𝑆 +

∑^𝐿_𝑙=1𝜑_𝑙𝑀𝑎𝑐𝑟𝑜_𝑐𝑜𝑛𝑡𝑟𝑜𝑙_𝑙,𝑡+ 𝑏𝑎𝑛𝑘_𝑖+ 𝜀_𝑖𝑡 (6) Because of the potential difference in bank size, profitability, capital adequacy between banks, the above estimation about the impact of MS may not only indicate the impact of the institutional difference between big 4 banks and the other banks. In order to exclude the impact of the different financial characteristics, we also consider the estimation by including bank financial indicators. Specifically, we include capital adequacy ratio (CAR), loan to deposit ratio (Ldratio), return of equity (ROE), net interest margin (NIM, defined by net interest income/interest earning assets), non-interest income ratio (NIIR, defined by non- interest income/total operating income). Zhu et al.(2016) suggested that ROE measures the whole profitability of banks for shareholders. Banks with lower ROE would have higher incentive to develop shadow banking business. Pan and Wei(2015), Guo and Zhao(2015) suggest that lower NIM means a weaker profitability on loans for banks, which led to a higher incentive for shadow banking activities. However, it might not be true in China. If banks do have a high NIM but face a loan quota restriction or higher spread, higher NIM may induce banks to increase off-balance lending through all kinds of shadow business.

NIIR indicates the profitability of banks through non-interest income or the dependency of traditional deposit and loan business of banks. Banks with higher NIIR might have more incentive to operate off-balance activities since these banks have relatively better profitability from non-interest business. However, when banks face tightened macro policy

18

including tightened credit quota, strict risky loan controls, higher spread, banks which was more dependent on traditional loan business, i.e., lower NIIR, would have higher incentive to operate shadow banking activities to earn from off-balance lending.

𝑆ℎ𝑎𝑑𝑜𝑤_𝑖𝑡 = 𝑐 + 𝑐₁𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡+ 𝑐₂𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡× MS + 𝑐₃𝑀𝑆 +

∑⁵_𝑖=1𝜗_𝑙𝐵𝑎𝑛𝑘𝐶ℎ𝑎𝑟_{𝑖,𝑡−1}+ ∑^𝐿_𝑙=1𝜑_𝑙𝑀𝑎𝑐𝑟𝑜_𝑐𝑜𝑛𝑡𝑟𝑜𝑙_𝑙,𝑡+ 𝑏𝑎𝑛𝑘_𝑖+ 𝜀_𝑖𝑡 (7) Table 4 shows the results of the equation (6) and (7). The column (1) is the result with ExM2 as the proxy of whole macro policy based on equation (6), and column (3) shows the result with the detailed macro policy shocks, including loan scale control shock(Exloan), safe-loan regulation shock(Reloans), interest rate constrain shock(Spread) and the net monetary policy shock (ExM2ext) based on equation (3). Column (2) and column (4) show the results with banks’ micro characteristics been included.

Table 4. Heterogeneous impact of macro policy on WMPs

(1) (2) (3) (4)

ExM2 -0.130^*** -0.084^**

(0.030) (0.036)

ExM2×MS -0.254^*** -0.194^***

(0.049) (0.047)

Exloan -0.076^** -0.071^*

(0.036) (0.039)

Exloan×MS -0.169^*** -0.152^***

(0.042) (0.037)

Reloans -0.397^*** -0.448^***

(0.062) (0.061)

Reloans×MS -0.030 -0.032

(0.033) (0.032)

Spread 0.184^*** 0.192^***

(0.049) (0.054)

Spread×MS -0.027 0.009

(0.055) (0.047)

ExM2ext 0.094^*** 0.114^***

(0.029) (0.032)

ExM2×MS 0.00008 -0.009

(0.044) (0.037)

MS -0.648 -0.437 -0.649 -0.657

(0.457) (0.496) (0.451) (0.478)

19

L.CAR -0.002 -0.021 -0.019 -0.047^***

(0.011) (0.014) (0.012) (0.011)

Hldr×L.Ldratio 0.011 0.019 0.002 0.003

(0.012) (0.013) (0.013) (0.012)

L.Ldratio -0.0001 -0.003 -0.009^** -0.011^**

(0.005) (0.004) (0.004) (0.004)

Hldr -0.735 -1.361 -0.153 -0.018

(0.899) (0.966) (0.974) (0.922)

Bank fixed effects Yes Yes Yes Yes

Macro controls Yes Yes Yes Yes

Micro controls No Yes No Yes

N 228 228 228 228

Standard errors in parentheses

* p < 0.1, ^** p < 0.05, ^*** p < 0.01

As can be seen from the column (1) of table 4, by using Exm2 capturing the whole monetary policy shock, we still find that tightened monetary policy stance would stimulate more WMPs, and vice versa. This effect is about twice higher for small banks compared to big 4 banks, even after the bank characteristics are controlled as shown in column (2).

When considering the detailed shocks, tightened loan scale shock would stimulate WMPs significantly, and this effect is more than twice higher for nonstate banks. For safe-loan regulation and interest rate constrain, tightened shocks would stimulate WMPs, however, this effect is not significantly different between big 4 banks and the others. For the net shock of monetary policy excluding the credit policy and interest rate constrain shocks, i.e., Exm2ext in our regression, the tightened monetary policy shock would actually constrain WMPs with no significantly different impact on nonstate and state banks. These heterogeneous impacts of the macro policy shocks are robust after the micro characteristics of banks are controlled. One standard deviation shock from safe-loan regulation has the largest impact on WMPs, compared to other standard deviation shocks, leading to an increase of WMPs by 40%. The second is the spread shock with an increase of WMPs by 19%, then the loan quota shock by 7% for big 4 banks and 22% for nonstate banks. One standard deviation of the net tightened monetary policy shock(ExM2ext) reduce WMPs by around 10%.

In summary, except the loan scale control policy, the other shocks have no significantly different impacts on big 4 and nonstate banks. Despite the direct control of big 4 banks by the central government, they also make regulatory arbitrage, especially when facing risky loan regulation and interest rate constrain.

20

3. The asymmetric impact of macro policy on WMPs

Since shadow banking business are from the macro policy regulatory arbitrage, it is natural to ask whether the impact of positive monetary policy shocks on WMPs is different with that of negative shocks. For example, banks would issue more shadow banking products when they face tightened monetary stance. Do they operate less shadow banking products when the monetary stance is loosened? Additionally, the macro policy has some different impacts on state banks and non-state banks based on the symmetric model, do small banks have different macro policy arbitrage patterns compared to big 4 banks when the macro policy shocks are decomposed into positive and negative parts? And would different macro policy tools have different asymmetric impacts? In this section, we decompose the monetary policy shocks into positive and negative part, then explore the possible asymmetric impact of macro policy shocks on WMPs. The model is as the following equation:

𝑆ℎ𝑎𝑑𝑜𝑤_𝑖𝑡 = 𝑐 + 𝑐₁𝑃𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡+ 𝑐₂𝑁𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡

+𝛽₁𝑃𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡× 𝑀𝑆 + 𝛽₂𝑁𝑀𝑎𝑐𝑟𝑜𝑝𝑜𝑙𝑖𝑐𝑦_𝑡× 𝑀𝑆

+ ∑⁵_𝑖=1𝜗_𝑙𝐵𝑎𝑛𝑘𝐶ℎ𝑎𝑟_{𝑖,𝑡−1}∑^𝐿_𝑙=1𝜑_𝑙𝑀𝑎𝑐𝑟𝑜_𝑐𝑜𝑛𝑡𝑟𝑜𝑙_𝑙,𝑡+ 𝑏𝑎𝑛𝑘_𝑖+ 𝜀_𝑖𝑡 (8) PMacropolicy and NMacropolicy are the positive and negative part of policy shocks respectively. As only one sample of Spread in our sample period is negative, we did not decompose the variable Spread and keep Spread in the regression. MS is the dummy variable as described before, which equals to 1 if the bank is not one of the big 4 state banks.

BankChar and Macro_control are defined as the same in equation (7).

The results are shown in table 5. Similar with the table 4, column (1) and (2) show the result as Exm2 being the proxy of the whole monetary stance, with and without micro bank characteristics controls respectively. Column (3) to (4) present the asymmetric impacts of those specific macro policy shocks, including loan quantity control, risky loan regulation and interest rate control and the net monetary policy shock.

As shown in column (1), positive monetary policy shock (ExM2>0), i.e., surprisingly loosen monetary policy stance, significantly reduce WMPs, with small banks shrinking WMPs 37% more than big 4 banks do. However, one standard deviation of negative monetary policy shocks(ExM2<0), i.e., unexpectedly tightened monetary policy shock can significantly reduce WMPs of big 4 banks by 9.5 percentages, while stimulate WMPs of small banks by 8.5%. However, this stimulating effect of negative shock disappeared and changed into constraining effect for small banks after the micro characteristics of banks are controlled as shown in column (2). One standard deviation of negative shock by ExM2 reduce WMPs by 13.8% for big 4 banks and 4.1% for small banks.

Column (3) and (4) show the results after decomposing the aggregate monetary policy shock into the detailed shocks. For the loan scale control, we can see the most similar impacts with the total aggregate monetary policy shock in column (2), just with a smaller extent. For the risky loan regulation, the positive shock, i.e., unexpectedly loosening regulation, will significantly reduce WMP of banks, and this effect is greater on small banks than on big 4 banks, which indicating that small banks are more independent on the risky loans. The negative shock, i.e., surprisingly stringent risky loan control, significantly stimulates banks to issue more WMP when the micro characteristics of banks are controlled,

21

and this stimulating effect is not significantly different for big 4 and other banks. For the net M2 growth shock (ExM2Ext), surprisingly monetary stance easing does not significantly impact the WMP issuance of big 4 banks. Negative monetary policy shock or unexpectedly tightened monetary policy can constrain WMP of all banks with a smaller impact on small banks. However, when the micro characteristics of banks are controlled, tightened monetary policy shock still reduce WMP of big 4 banks, but insignificantly. The negative shock stimulates small banks to issue more WMP at the 10% significant level after controlling the financial characteristics of banks.

Table 5. Asymmetric impact of macro policy on WMPs

(1) (2) (3) (4)

P.ExM2 -0.296^*** -0.265^***

(0.050) (0.043)

N.ExM2 0.0951^** 0.138^***

(0.043) (0.036)

MS×P.ExM2 -0.110^** -0.101^**

(0.054) (0.045)

MS×N.ExM2 -0.180^*** -0.0966^**

(0.054) (0.046)

P.Exloan -0.168^*** -0.213^***

(0.039) (0.046)

N.Exloan 0.0589^* 0.127^***

(0.036) (0.044)

MS×P.Exloan -0.0564 -0.0724^*

(0.051) (0.041)

MS×N.Exloan -0.0891^** -0.0420

(0.039) (0.031)

P.Reloans -0.359^*** -0.416^***

(0.074) (0.076)

N.Reloans -0.0857 -0.300^***

(0.083) (0.090)

MS×P.Reloans -0.0782^** -0.0933^***

(0.031) (0.028)

MS×N.Reloans 0.0163 0.0175

(0.031) (0.025)

L.ROE×P.Reloans 0.0215^*** 0.0197^***

(0.003) (0.003)

L.ROE×N.Reloans 0.00312 0.00989^**

(0.004) (0.004)

Spread 0.172^*** 0.187^***

(0.043) (0.050)

MS×Spread 0.000298 -0.00299

(0.060) (0.048)