key: cord-0754836-d2dvdzb1 authors: Zhang, Xiaoqian; Wang, Zhiwei title: Marketization vs. market chase: Insights from implicit government guarantees date: 2020-06-27 journal: nan DOI: 10.1016/j.iref.2020.06.021 sha: a4912258d669ec0f1373452d1eebe02d920967f4 doc_id: 754836 cord_uid: d2dvdzb1 Abstract Local Government Financing Vehicles (LGFVs) and state-owned enterprises (SOEs) provide implicit guarantee during the issuing of bonds, thereby reducing their funding cost. The credit spreads are lower when issued by a LGFV with a higher administrative level. This means that implicit guarantee is also strengthened with government centralization. We also explain the anomaly of municipal corporate bonds (MCBs)’ spreads decrease after a marketization regulation of removing implicit guarantees. This paper provides strong evidence that the market will chase implicit guarantee when default wave comes even under tight government regulations. Implicit government guarantees that favor state-owned firms are becoming more prominent during recessions. It reduces the financing cost, and distorts corporate investment decisions. Current loan-level data. In the post-stimulus period, Chinese local governments resorted to non-bank debt with the development of the Chinese corporate bond market (Chen et al., 2020b) . Both papers demonstrate the effect of government guarantee on firm financing. The difference between government guarantees between China and the U.S. is highlighted in municipal bonds. In China, municipal bonds are also issued by Local Government Financing Vehicles ( LGFVs), which are state-owned enterprises to support the infrastructure investment. There are two areas in particular that make China's municipal corporate bonds are different from American municipal bonds from the following two perspectives. First, individual investors hold municipal bonds directly or through municipal bond funds in the U.S., while financial institutions buy the municipal corporate bonds (MCBs) in China. Second, the credit rates of China's MCBs are highly and have never defaulted. As our data shows, almost all the MCBs were high rated in China while there are lots of speculative-grade municipal bonds in the U.S. as Babina et al. (2020) show. Borisova et al. (2015) find that government ownership is generally associated with a higher cost of debt in 43 countries over 1991-2010. In contrast, we find China market chases more implicit government guarantees. This paper provides strong evidence that the implicit government guarantee does exist and -3-reduces the funding costs by exploring bonds' credit spreads. We find that the spreads of MCB and other local SOEs are 85 and 81 basis points lower than that those of POE bonds while their ROA is significantly lower. Central SOEs have even lower spreads. The implicit government guarantee does exist in MCBs when we control bonds' credit rates. Our finding stems from a deeper analysis from three perspectives. First, MCBs and other local SOEs have about 1% lower financing cost when we introduce bond characters and regional economic development. This reveals that the main concern is that of misallocation of capital as Gertler and Klenow (2019) addressed. Government credit may cause crowding out of private companies in the same industry (Ru, 2018) . Second, we further exploit the effect of administrative levels of MCBs. Our results identify political decentralization (Bardhan, 2016) in different administrative level of LGFVs. Administrative level decentralization increases funding costs, which is consistent with Huang et al. (2017) . Third, our in-depth research on guarantees reveals the heterogeneities of implicit and explicit government guarantees. We find that credit spreads are about 12 bps higher with guarantors or explicit government guarantees, while implicit guarantee significantly decreases the spreads of bonds by 1%, very significantly. The mechanism is adapted from regional government intervention and macro heterogeneities. We use two indicators to identify government intervention following Hao and Lu (2018) , based on employees or and GDP, respectively. We find that there are significant effects stemming from government intervention. Regional macro heterogeneities are also examined from regional economic development, fiscal statement and bank loan credit. We find that the implicit government guarantee has more power over bank loans and decreasing the MCB's financing cost in more developed regions and more bank loans. Economic development dominates this effect. In order to handle the endogeneity, we introduce an instrumental variable to predict local GDP per capita, an intersection of regional -4-medical and health expenditure in public expenditure per capita and the pandemic. The latter is a dummy variable, displaying either high or low cumulative confirmed cases of coronavirus in this region. The results are robust. The reason why we introduce this IV based on public health is on the basis of the power of local governments pertaining to public health, as pointed out by journalists (for example, the lead article on Economist May 30, 2020 2 ). Prominent studies have a tendency to focus on the black side of government intervention, like conflicts among different parties in production under asymmetric information (Cailaud, et al., 1988) , policy burdens and soft budget constraint (Lin & Tan, 1999) , making inefficient investment decisions (Chen, et al., 2011) . 3 However, price is endogenous to government policy, which is a clear sign of a key economic force. When governments make decisions based on information they learn from market prices, this affects the amount of information the government can obtain. Bond and Goldstein (2015) documents that the government discloses details about a variable that would be beneficial to speculators. This in turn, helps the government due to the reduction of risk faced by speculators because of the disclosed stock prices. But Bond and Goldstein (2015) maintains the concern that the government would conduct major interventions without having precise information about the costs and benefits of doing so. China's bond market and the recent tightening of interventions provide insights of Bond and Goldstein's concern. China government has much more information on Municipal Corporate Bonds (MCBs) than the speculators, but it kept marketizing MCBs without learning from the information on MCBs' price. The implicit government guarantee should decrease as these marketing policies being practiced, in contrast, we find that market kept chasing these MCBs because of the implicit guarantees which they regard as much safer. Our paper is related to the literature on political connections or state ownership. Previous research focused on the political connections, originating from the findings of Fisman (2001) . Fisman and Wang (2015) which provide evidence on how political connections enable firms to avoid compliance measures based on their finding that the death rate of workers for connected companies is 2~3 times that of unconnected firms. Gao et al. (2019) recently find that established local politicians engage less often in selective defaults on bank loans. Our paper provides a perspective from the bond market to explain the bargaining between government and market. It also provides evidence on the research of state ownership. Shi and Zhang (2018) institutions have entered the Chinese bond market, with a debt holding amount of more than 2.1 trillion RMB. We provide an insight from implicit government guarantee on an aspect of the post-stimulus period in China. While our paper draws evidence from China, the insight is also able to explain the relationship between government and market more broadly. This paper may also provide new evidence on some important theoretic literature, such as Hart and Zingales (2011) , Bond and Goldstein (2015) , who call for the state to make use of information contained in market prices, as many researchers and policy makers suggest. The paper is organized as follows. Section 2 briefly reviews China bond market and regulations, and the implicit government guarantees implied in MCBs. Section 3 describes the sample and data. Section 4 presents the empirical results. Section 5 analyzes the mechanism and section 6 further exploits the bargaining between state and market. Section 7 introduces Bonferroni tests, extreme bounds analysis and placebo tests as robustness checks. Section 8 summarizes our findings and has bypassed its own stock market capitalization and the level of the U.S. bond market in 2017. Therefore, the Chinese bond market plays an increasingly important role, whether within its own capital market or the global perspective. Panel B of Figure 1 5 compares the proportions of government debt over GDP in G20 countries. China has a much lower government debt ratio than other developed countries except Australia, and even some developing countries like India and Brazil. The reason may lie in that some Chinese government debts have been converted into corporate debt, municipal corporate bonds (MCB), whose issuer is a special corporation called as Local Government Financial Vehicle. It is more convenient for the government to conduct financing and investment through these MCBs. Chinese government debt -8-is implied within corporate debt, which is a unique perspective of focusing on the bargain between state and market. [Insert Figure 1 ] Jiang et al. (2020) provide an overview of Chinese capital markets before 2015. They documented that despite the accelerated growth of China's bond markets are growing fast but bank financing still dominates debt financing in China. As Panel A of Figure 1 shows, Chinese bond market grows rapidly after 2015, the period examined by Jiang et al. (2020) Based on issuing entities, the market is classified into three broad bond categories; government, financial, and corporate bonds. According to Amstad and He (2019) , government bonds is 57.55% in China which is lower than US (63.94%). Why does China issue less government bonds than the U.S.? The reason Amstad and He (2019) did not mention is that Chinese local government obtains finance through MCBs which is issued by LGFVs. This means, more and more government debts are regarded as corporate bonds which embedded in national balance sheet. LGFVs. LGFVs played an integral role in implementing the fiscal expansion of 2009 and 2010. The local government could not issue bond to raise money and is heavily dependent on the central government before 2015. The local government has to establish LGFVs to raise money indirectly by issuing MCBs. Bai et al. (2016) and Chen et al. (2020a) This paper focuses on an aspect of post-stimulus period in China. In this period, the government tightened the regulations on LGFVs. Figure 2 shows the revolution of the tightening regulation on China's LGFVs. As the non-standard municipal corporate crisis happened in 2011, and the first SOE bond Baoding Tianwei defaulted in April 2015, the market began to explore the value of implicit government guarantee (Jin et al. 2018 process, market is still MCBs as a safe-haven asset due to its implicit guarantee nature. [Insert Figure 3] Implicit government guarantee, also known as soft budget constraints, is implied in almost all the developed and developing countries. Dewatripont and Maskin (1995) documents that centralized economy is more likely to have soft budget constraints. The existence of implicit guarantee requires the government to spend more energy to manage the financial security problems caused by this implicit debt, thus triggering the credit crisis of local governments. Enterprises will blindly expand aggressive investment activities with the guarantee, constantly expand credit scale, increase leverage ratio and aggravate risk problems. The market's general expectation of soft budget constraint will disrupt the normal development of financial market order. Because of their close relationship with the government, SOEs are strongly motivated to seek help from the government in the form of soft budget -12-constraints (Lin and Tan, 1999) . Banks will tighten credit to higher-risk borrowers with less collateral and higher regulatory costs, which result in credit rationing (Stiglitz & Weiss, 1981) . Conversely, if banks prioritize the allocation of credit to politically connected customers, such as SOEs, it is because they often perpetuate implicit or explicit government guarantees. Huang et al. (2018) measures the risk of implicit guarantees on Chinese shadow interbank products. They find that banks extend more implicit guarantees to shadow bank products when their solvency deteriorates. (2017) verifies that political connections offset the negative relation between investment and political uncertainty because it reduced information asymmetry for connected firms resulting in a delaying of investment in anticipation of future lucrative tax incentives. Nagano (2018) documents that firm information asymmetry, a common feature of emerging markets, moderates the negative relationship between the firm's financial constraints and debt security issuance. The issuing of credit spreads are more dependent on the fluctuations in the macroeconomic cycle (Gilchrist and Zakraisek, 2012), as well as the quality of issuer (Benzion et al., 2018) We start with a sample of all Chinese enterprise bonds (EB hereafter), exchange-traded corporate bonds (ETCB hereafter), and medium-term notes (MTN hereafter) issued between 2010 and 2018. Amstad and He (2019) We only keep state-owned enterprises (SOEs) or private-owned enterprises (POEs), thus 11740 bonds constitute our final sample. [Insert Table 1] We collect bond characteristics and financial indicators of issuers from WIND database. Bonds issuers are required to submit quarterly financial reports by Securities Law of the People's Republic of China. Therefore, although 85% of bond issuers are not listed, their financial indicators are able to be obtained. We use the data from their financial reports in the last quarter to reduce the concern of endogeneity. All corporate-level financial indicators have been winsorized by year. The definitions of variables are shown in Appendix Table A1 . Bond daily transaction data and macro-economic data are obtained from CSMAR database. We also use an IV based on the recent shock to the global economy, COVID-19, to reduce the concern from endogeneity, because pandemics are more exogenous compared with policy shocks. We use credit spread to measure the financing cost of the bonds. Spread is the difference between bond yield and the matching central government bond yield, which has the same cash flow characteristics with the same issuance date and maturity following Ang et al. (2019) . We also use the second index, SpreadCDB, to measure credit spread, defined as the bond yield minus the matching China Development Bank (CDB) bond yields following Chen et al. (2020a) . The results are robust. Panel A of Table 2 compares MCBs and other SOE bonds with POE bonds in order to reveal the existence of implicit government guarantee. POEs are regarded as being without government guarantee, therefore we use POEs as the basis group. The comparison in Panel A of Table 2 demonstrates the existence of government implicit guarantee in MCBs, central SOEs and local SOEs. [Insert Table 2 ] Table 3 reports the average credit spreads of the bonds according to administrative regions year by year. We divide the sample into ten subsamples by sorting deciles of the regional GDP per capital, with the lowest group denoted as 1 and the highest group denoted as 10. The result implies that bond spread is relatively lower in the developed regions with higher average GDP in average. Among them, the anomaly that only exists in 2010 may be due to the relatively small number of bonds issued during -17-the year, which was greatly affected by extreme values. We will further discuss the impact of macro variables on bond spreads in Section 5.2. We also divide the sample by sorting city-level GDP in Panel B of Table 3 . The result is consistent with Panel A of Table 3 . [Insert Table 3] 4． ． ． ．Empirical Results First, we propose the following benchmark model to test hypothesis 1: includes 5 variables of the firm issuing the bond; return on assets (ROA), total debt over total asset (Leverage), the difference of current operating income and initial operating income over initial operating income (Salegrowth), the logarithm of total asset (FirmSize), and the difference between the -18-year firm established and year bond issuance (Age). α s, α p and α t is industrial, provincial and year fixed effect, respectively. Table 4 shows the regression results. Column (1) is the baseline model following Ang et al. (2019). Following Erel et al. (2015) , we introduce two impact factors into the regression in column (2), Privatesector controlling the development of POEs, and MarketCap controlling the financing from equity market in province p. Column (3) implies that the financing cost of SOE is 1.004% lower than that of POE. Column (4)-column (6) show the same results that central SOE, local SOE and MCB have 1.259%, 0.984% and 0.967% lower than POE ceteris paribus. We can find that the coefficient of central SOEs is even lower than local SOEs, which implies that central SOEs are provided a stronger implicit guarantee from the central government. In column (6), we define MCB i as Treated i , which equals to 1 when the bond issuer is LGFV, while it equals to 0 when the issuer is POE. The result shows that mean spread of MCBs is lower than that of POEs by 0.967%. [Insert Table 4] To deal with the concern that MCBs usually have higher credit ratings compared to bonds of POEs, we further compare the matched sample in Table 5 . Column (1) and column (3) report the regression results in non-matched sample. Column (2) and column (4) report the results in propensity-score matched sample with the same region, the same year and the same credit rate where we select the closest propensity score without replacement. The results are consistent. The coefficients of MCBvsPOE are all significantly negative. LGFV, MCB's issuer, has a significantly lower financing cost than POEs, around 90 basis points. [Insert LGFV and the level of LGFV is prefectural level, while it equals to 0 when the issuer is POE. CountyLevel is a dummy variable. It equals to 1 if the issuer is LGFV and the level is county and county-level cities, while it equals to 0 when the issuer is POE. (3) and column (4) show. County-level LGFVs have around 0.83% lower spreads than POEs as the last two columns demonstrate. Our results imply that the provincial government provides the strongest implicit guarantee, followed by the prefecture-level government and county level. We control the credit rating fixed effect in column (2), column (4) and column (6), which are consistent with column (1), column (3) and column (5). Our paper provides strong evidence on Political Decentralization as mentioned by Bardhan (2016) and Huang et al. (2017) . China state (the central government) has been decentralizing those SOEs with poor performance. Our paper provides evidence to their findings by the side of MCBs. [Insert Table 6] 4.3. Implicit guarantee compared with explicit guarantee Some credit bonds have guarantor to increase market confidence in anticipation of successful issuance. However, the market will respond differently to the guarantor's qualifications. Then we study the impact of different kind guarantor. where Guarantee i is a dummy variable to indicate the guarantee state of bond i. We use three variables, DummyGuarantee, ExGovGuarantee and ImGovGuarantee, respectively. The first indicator, DummyGuarantee, equals 1 if the bond is guaranteed, otherwise is 0. The coefficient of DummyGuarantee in column (1) in Table 7 is positive, which means the spread of guaranteed bonds is 0.241% higher than those unguaranteed. The second indicator, ExGovGuarantee, is a dummy variable to indicate explicit government guarantee, which equals 1 if the bond is guaranteed by government or SOE, otherwise is 0. The result in column (2) shows that the credit spread is 0.156% higher. The third indicator, ImGovGuarantee, is a dummy variable to grasp the effect from implicit government guarantee. We only keep the bonds unguaranteed. It equals to 1 if the bond is unguaranteed but the issuer is state owned, otherwise 0. This is because unguaranteed SOE bonds are largely subject to implicit government guarantees for political purposes to prevent bond defaults. Column (3) shows that the coefficient of ImGovGuarantee is significantly negative, which proves that the spread of state-owned corporate bonds with implicit government guarantees is averagely 1.074% lower. We further introduce the intersection terms with MCBvsPOE to ascertain the heterogeneity between MCBs and POEs in column (4)-column (6). The ImGovGuarantee in column (1)-column (3) equals MCBvsPOE in column (4) -column (6). Therefore, these three regressions run performance tests between the two explicit guarantees and implicit guarantee. The regression results show that both the guarantees are significant. Implicit guarantee, which is indicated definitely as MCBsvsPOE, reduces the spread about 1%, i.e. 100 basis points. For the two explicit guarantees, DummyGuarantee and ExGovGuarantee, both reduce the spread for the control group, POEs. The two intersection terms are significantly positive and more than the coefficients of the two explicit guarantees, implying that explicit guarantee increases MCB's spread. Therefore, the conclusion from column (4)-column (6) is much clearer than the above conclusion in column (1)-column (3) in which MCB and other bonds of SOEs are much more than POEs. Considering the heterogeneity between MCB and POE, we find that for POEs, explicit guarantee has a significant effect of reducing the spread, but it increases the spread of MCBs because the of market's concerns. [Insert Table 7 ] This section aims to explain why there are variances in spread in different regions as Table 3 shows. First, we exploit the effect of government intervention by the following model, where GovInt p,t-1 is the indicator measuring the strength of government intervention in province p. We use two measures following Hao and Lu (2018) , the ratio of the number of employees in public administration, social security and social organizations to the total number of employees (GovInt1) and the ratio of government expenditure to local GDP (GovInt2). Table 8 reports the result of equation (4). Column (1)-column (2) compares MCB with POEs. Column (3)-column (4) compares MCB with local SOEs. Both of the coefficients of government intervention based on GDP, GovInt2, are significantly positive, which is helpful to explain our findings of Table 3 . There is a relatively low degree of marketization in the provinces with strong government intervention, that goes against the firm's objective of maximizing profit. The government has been interfering with the companies' normal operations and risk is increasing, so the credit spread increases. The intersection terms exploit the heterogeneity. Column (1) and (2) compares the difference in mechanism of government intervention's impact on spreads between MCBs and POE bonds. Column (3) and (4) compares the difference in mechanism between MCBs and local SOE bonds. The coefficients of intersection terms are all insignificant, implying the robustness of our conclusion. [Insert Table 8 ] In this section, we discuss about the effects of different macro environments: Spread i =β 1 Fiscal r,t-1 +γ 2 Bond i +γ 3 Issuer j,t-1 +α s +α t +ε i where Fiscal represents the provincial fiscal statement in the region r where the issuer is located. Here we consider both provincial-level and city-level fiscal statements. Provincial-level variables include -23-the logarithm of the per capita GDP (LnperGDP p,t-1 ), the logarithm of the local GDP of the real estate sector (RealEstateGDP p,t-1 ), the logarithm of the local tax revenue of the real estate sector (RealEstateTax p,t-1 ) and the growth rate of housing prices in each province (HousePriceGrowth p,t-1 ). City-level variables include the logarithm of per capita GDP of the city of the LGFV (LnpercityGDP c,t-1 ) and which is defined as the value of the outstanding loan balance of a financial institution prefectural-level, financial institution divided by the GDP of that city (PrivateCredit c,t-1 ). Since Macro includes the provincial-level variables, we exclude these two macro control variables to reduce multicollinearity. We use MCBs as the sample to test equation (5) in Table 9 because we only focus on the impact of government fiscal situation on MCBs, which have close relationship with the government. Column (1)-column (6) ascertains the effect from each provincial-level fiscal statement. Column (7) considers their effects together. First, from the effect of local GDP, column (1), column (7) and column (8) all imply that spreads decline in developed province or city. Column (7) shows the impact from GDP is the strongest among all the factors. This is consistent with our prior findings. MCB has about 0.395% decrease in spread with every 1% increase in local GDP per capita, which is even stronger in the city level. Second, we also consider the impact from real estate following Ang et al. (2019) . We use three variables to indicate the impact from real estate, the local GDP of the real estate sector, real estate tax and the growth rate of housing prices. The regression results show that when the real estate GDP increases by 1%, the issued credit spread will decrease by 0.161%, and significantly at the 1% level. The results in columns (5) and (6) show that the greater the degree of government intervention, the lower the spread of the MCBs. By adding all provincial variables in column (7), LnperGDP p,t-1 and RealEstateGDP p,t-1 still exert a significant negative impact. This implies that MCBs have lower financing cost in developed province. Third, further considering the impact from bank loans, we keep the strongest impact factor, GDP but at city level, and add PrivateCredit, the ratio of bank loans over GDP in the city. Column (8)-column (10) show significant impact from these two city-level factors. The coefficients are both negative whenever we put them in the regression separately or together. GDP still has significantly negative impact. For bank loan's effect, the higher PrivateCredit means it is easier for the firm to obtain bank loans. Our results show that if PrivateCredit increases by 1%, MCB's spread falls 0.138%. But the effect shrinks rapidly, by 0.058% if we consider the effect of economic development as column (10) shows. This provides evidence that economic development has stronger effect on reducing MCB's financing cost than bank loans do. [Insert Table 9 ] In order to reduce the endogeneity, we also introduce an exotic instrumental variable to predict lnperGDP. Since we use GDP per capital, the population characteristic and investment should be critical to it. The COVID-19 outbreak is an exogenous shock to the global economy, as recent papers examined the relationship between the pandemic and economic development, ex. Jia et al. (2020), Fang et al. (2020) and Qiu et al. (2020) . We introduce an intersection of regional medical and health expenditure in public expenditure per capita with pandemic, and lag item of regional GDP in the first stage to predict regional economic development 6 . The 2SLS regression is as follows, lnperGDP rt =β 1 HealthExpenditure , × Pandemic+ lnperGDP ,t-1 +α r r r r +α t +ε (6) where HealthExpenditure is the logarithm of regional medical and health expenditure in public expenditure per capita. Pandemic is a dummy, which equals 1 above the median of the cumulative confirmed cases in this region on April 8, 2020, otherwise is 0. April 8, 2020 is the date when Wuhan, the city that was the epicenter of the coronavirus outbreak, reopened and the lockdown restrictions were eased, indicating that the number of cases has been steady afterwards. We also use the number of deaths and recoveries as well. The results are robust. Table 10 still reports the robust results as Table 9 shows. [Insert year and same province. [Insert Figure 4 ] We use DID model to exploit the structural change of Article 43 as follows, where Post t is a dummy, equaling to 1 if the bond was issued after 28 September, 2014. α s means the industry fixed effect, and α p means province fixed effect and α t shows year fixed effect. Column (2) and (4) show the result of PSM-DID, which we match the bond credit rating, year and province fixed-effect. Column (2) represents that compare to the local SOEs' bonds with same credit rating and issuance year and province. (1) shows. The decrease is even lower when we use the matched sample, even 0.28% lower than local SOE. Column (3) and column (4) show the results compared with POE. The coefficients of MCB and intersection term, MCB×Post, are significantly negative, which implies that MCB has lower spread than POEs by around 0.3% before the shock, then even lower after that. MCB has even lower financing costs by about 1% whether we use matched sample or not. [Insert Table 11 ] Table 12 further exploits the impact from administrative level before and after Article 43 by considering capital-level and prefectural-level MCBs, respectively. For the capital-level MCBs, we divide the samples into ten groups according to provincial GDP where the issuer of MCB located in. The group of highest GDP is denoted as High GDP, and the lowest is denoted as Low GDP. As for city level MCBs, we divided the samples into ten groups according to GDP of the city where the -27-issuer of MCBs located. Table 13 shows that the mean spread of each region is lower after the shock than before. For the capital-level MCBs, we find that MCB in high GDP provinces has 0.504% lower spread than in low GDP provinces before Article 43 in column (1). This difference is even strengthened after the shock, which is 0.568% as column (2) shows. For prefectural level MCBs, we still find that this difference is significant, 0.222% in average before the shock. But after the shock, there is no significant difference between high and low cities among the prefectural MCBs. Thus, we find that the mean spread of MCBs even declined after the implementation Article 43 for MCBs at both capital and prefectural level. [Insert Table 12 ] Finally, we will explain why the spread declines even after the government aims to begin marketization of those MCBs. The reason lies in the market concern on bond defaults. We introduce the variable Default to indicate our hypothesis. The regression model is as follows, where we use the total default, local SOE default and POE default, respectively. (4) considers these two factors together, the conclusion still holds and we find POEs default has more impact. Those POEs' defaults restore the market's confidence in the MCBs. That is the reason why LGFV could issue bonds at lower spreads. [Insert Table 13 ] We calculate the Bonferroni's upper bound of joint distribution on the true joint p-value. We have 13 controls and an independent variable in the base model, so we use the one-thirteenth p-value (0.01/13) as the significant standard. In Panel A of Table 14 , ***, **, * means p<0.00077, p<0.0038, p<0.0077 respectively, and the results still are robust when we use the strict Bonferroni bound. The result is consistent with our benchmark model of Table 4 . [Insert Table 14 ] Leamer (1983) We find strong evidence that government guarantee reduces the funding cost by examining the credit spread of corporate bonds in China. MCBs have significantly lower funding costs because of the implicit government guarantee. According to administrative level of LGFVs, states provide the strongest implicit guarantee, then prefectural level, and county level is the last. We also find the heterogeneity of explicit guarantee and implicit guarantee between MCBs and POEs. Although explicit guarantee reduces POE bond's spread, it is a bad signal for the market on MCBs where the spread increases if it is issued with explicit guarantees. In contrast to explicit guarantee, implicit guarantee is good news because the market has more confidence on MCBs rather than POE bonds. We then analyze the mechanism by two perspectives, government intervention and macro heterogeneities. We find that MCB still has lower spread than POE bonds even introducing two indicators of government intervention. Regional economic development has the strongest effect whenever we use province-level or city-level GDP per capita. Bank loans reduce the spread too, but this effect shrinks rapidly introducing economic development. These results contribute to the literature on policy decentralization and government intervention. Furthermore, the bargaining between market and government is exploited, by introducing the Indicators of government intervention. Divide government expenditure by local government GDP (Hao & Lu, 2018 ) GovInt2 Indicators of government intervention. Divide the number of employees in public administration, social security and social organizations by the total number of employees (Hao & Lu, 2018 ) PrivateSector The total employees in private and self-employed enterprises over total employees in each province MarketCap The market value of listed companies over GDP (Erel et al., 2015) in each province LnperGDP The logarithm of GDP per capital in each province RealEstateGDP The logarithm of GDP of real estate sector in each province RealEstateTax The logarithm of local tax revenue of real estate sector in each province HousePriceGrowth Growth rate of housing prices in each province LnpercityGDP The logarithm of per capita GDP in each city PrivateCredit The total outstanding loan over GDP (Erel et al., 2015) in each city HealthExpenditure The logarithm of regional medical and health expenditure in public expenditure per capita. A dummy equals 1 above the median of the cumulative confirmed cases in this region on April 8, 2020, otherwise is 0. April 8, 2020 is the date of the end of the lockdown and the reopening of the city of Wuhan.. Municipal Corporate Bond dummy variable. LGFV is 1, and POE is 0 RateAAA Dummy variable. It equals to 1 if the bond is AAA-rated when issued, otherwise is 0. lnBondSize The logarithm of bond size (in million RMB) lnMaturity The logarithm of bonds maturity (in year) EB Dummy variables. It equals to 1 if the bond is enterprise bond regulated by NDRC, otherwise is 0. Dummy variables. It equals to 1 if the bond is middle-term note, otherwise is 0. Dummy variable. It equals to 1 if the administrative level is province, provincial capitals and cities specifically designated in the state plan (Single-listed-city), otherwise is 0. PrefecturalLevel Dummy variable. It equals to 1 if the level is Prefecture-level city, otherwise is 0. Asset-liability ratio (%) Salegrowth Current operating income minus initial operating income over initial operating income FirmSize The logarithm of total asset of the issuer. The difference between the year firm established and year bond issuance Figure 1 (a) reports the growth of the Chinese bond market capitalization scaled by GDP (in bars) or stock market capitalization (in lines). As for the proportions of government debt over GDP in G20 countries, which implies that China has a much lower government debt ratio than developed countries except Australia, and even some developing countries like India and Brazil. Figure 2 shows the revolution of the tightening regulation on China's LGFVs. As the non-standard municipal corporate crisis happened in 2011, and the first SOE bond Baoding Tianwei defaulted in April 2015, the market began to explore the value of government's implicit guarantee (Jin et al. 2018) . Under the background of deinventory and deleverage, the secure payout of state-owned enterprises is broken, resulting in the market turmoil. Implicit guarantee should have been gradually losing its effectiveness. On the other hand, the rapidly increasing implicit debt of local government has drowned central government's attention, and a growing number of MCBs improve the debt risk. The State Council issued Article 43 request LGFV to be decoupled from local government and the debt cannot be regarded as implicit debt of local government, which decrease the implicit guarantee of local government. Article 463 regulates on local government financing. Article 43 regulates the LGFV and clearly defining the boundaries between the government and enterprises. The first wave of bond defaults broke out. Table 1 Summary statistics of China's bond market data. Non-MCBs but SOE Bonds Bonds of POEs Year ETCB EB MTN ETCB EB MTN ETCB EB MTN 2010 1 18 1 3 20 0 0 2 0 2011 0 11 1 5 25 1 1 2 0 2012 7 141 12 17 85 9 5 12 0 2013 13 357 17 3 87 19 7 16 0 2014 12 620 98 5 110 76 8 7 2 2015 105 353 214 133 160 285 88 13 20 Table 2 compares MCBs and other SOE bonds with POE bonds in order to reveal the existence of implicit government guarantee. POEs are regarded as being without government guarantee, therefore we use POEs as the basis group. Columns (1) to (4) report the mean and median credit spreads. Medians are in the brackets. Columns (5) to (7) show the difference compared to POEs. The significance of mean or median tests is based on one-side t-tests or rank-sum tests (in parentheses). The comparison in Table 2 In Table 3 , we summarize the spreads according to the administrative regions year by year since bond spreads are affected by regional characteristics. We divide ten subsamples according to deciles of province GDP, and the lowest group is denoted as 1, while the highest group is denoted as 10. The result shows that bond spreads are relatively small in provinces with higher GDP and relatively larger in provinces with lower GDP on the whole. We will further discuss the impact of macro variables on bond spreads in Section 5.2. We also divide ten subsamples by city GDP in Panel B of Table 3 , and the result is consistent with Panel A of Table 3 . The Effect of Implicit Guarantee (1) (2) (3) (4) (5) (6) Table 4 shows the regression results. See Appendix Table A for definition of the variables. *** p<0.01, ** p<0.05, * p<0.1. To deal with the concern that MCBs usually have higher credit ratings compared to bonds of POEs, we further compare the matched sample in Table 5 . Column (1) and column (3) reports the regression results in non-matched sample. Column (2) and column (4) reports the results in propensity-score matched sample with the same region, the same year and the same credit rate where we select the closest propensity score without replacement. See Appendix Table A for definition of the variables *** p<0.01, ** p<0.05, * p<0.1. Table 6 shows MCBs are lower in different administrative level compared to POEs. We use capital-level LGFVs in Column (1) and column (2), and we use prefectural-level LGFVs to compare against POEs in column (3) and column (4). County-level LGFVs in the last two columns. We control the credit rating fixed effect in column (2), column (4) and column (6), which are consistent with column (1), column (3) and column (5). See Appendix Table A for definition of the variables .*** p<0.01, ** p<0.05, * p<0.1. Table 7 shows the different results among variety of guarantee. See Appendix Table A for definition of the variables. *** p<0.01, ** p<0.05, * p<0.1. Table 8 reports the result of equation (4). Column (1) and (2) compare the difference in mechanism of government intervention's impact on spreads between MCBs and POE bonds. Column (3) and (4) compare the difference in mechanism between MCBs and local SOE bonds. See Appendix Table A for definition of the variables.*** p<0.01, ** p<0.05, * p<0.1. We use MCBs as the sample to test equation (5) in Table 9 because we only focus on the impact of government fiscal situation on MCBs, which have close relationship with government. Column (1)-column (6) exploit the effect from each provincial-level fiscal statement. Column (7) considers their effects together. Column (8)-column (10) show significant impact from these two city-level factors. For bank loan's effect, the higher PrivateCredite means it is easier for the firm to obtain bank loans. See Appendix Table A for definition of the variables.*** p<0.01, ** p<0.05, * p<0.1. (1) (2) (3) Step 1: lnperGDP Step 2: Spread Step 2 In order to reduce the endogeneity, we also introduce an exotic instrumental variable to predict lnperGDP. The COVID-19 outbreak is an exogenous shock to the global economy, as recent papers examined the relationship between pandemic and economic development, ex. Jia et al. (2020) , Fang et al. (2020) and Qiu et al. (2020) . We introduce an intersection of regional medical and health expenditure in public expenditure per capita with the pandemic, and lag item of regional GDP in the first stage to predict regional economic development. We also use the number of death and recoveries. The results are robust. See Appendix Table A for definition of the variables.*** p<0.01, ** p<0.05, * p<0.1. The shock of Article 43. (1) (2) Table 11 shows the heterogeneous effects of MCB with local SOE and POE, and the structural change from the shock of Article 43. We use the unmatched sample in column (1) and matched sample in column (2). The regression results are consistent. Column (3) and column (4) show the results compared with POE. See Appendix Table A for definition of the variables.*** p<0.01, ** p<0.05, * p<0.1. High-Low -0.504*** -0.568*** -0.222** -0.154 Table 12 further exploits the impact from administrative level before and after Article 43 by considering capital-level and prefectural-level MCBs, respectively. For the capital-level MCBs, we divide the samples into ten groups according to provincial GDP where the issuer of MCB located in. The group of highest GDP is denoted as High GDP, and the lowest is denoted as Low GDP. As for city level MCBs, we divided the samples into ten groups according to GDP of the city where the issuer of MCBs located. See Appendix Table A for definition of the variables.*** p<0.01, ** p<0.05, * p<0.1. Impact of bond defaults. (1) (2) (3) (4) Table 13 reports the regression results, where we use the logarithm of number of defaulted bonds, we also use the logarithm of value of the defaulted bonds and the cumulative number of defaulted bonds which are all consistent. We still focus on the MCB subsample in the following regressions. Column (1) shows that the greater the number of default bonds, the lower spread of MCB has. Column (2) and column (3) further consider the impact from local SOE's default and POE's default. Column (4) considers these two factors together. See Appendix Table A for definition of the variables.*** p<0.01, ** p<0.05, * p<0.1. Chinese bond market and interbank Market. Chapter for the Handbook of "China's Financial System The great wall of debt: Real estate, political risk and Chinese local government financing cost Heterogeneous Taxes and Limited Risk Sharing: Evidence from Municipal Bonds. The Review of Financial Studies The long shadow of China's fiscal expansion State and Development: The need for a reappraisal of the current literature Debt composition and lax screening in the corporate bond market Government intervention and information aggregation by prices Government ownership and the cost of debt: Evidence from government investments in publicly traded firms Government intervention in production and incentives theory: a review of recent contributions Pledgeability and asset prices: evidence from the Chinese corporate bond markets Quantifying liquidity and default risks of corporate bonds over the business cycle Government intervention and investment efficiency: Evidence from China The financing of local government in China: stimulus loan wanes and shadow banking waxes Credit and efficiency in centralized and decentralized economies Credit-Induced Boom and Bust Do Acquisitions Relieve Target Firms' Financial Constraints? Politically Connected Firms Public governance and corporate finance: Evidence from corruption cases Human Mobility Restrictions and the Spread of the Novel Coronavirus (2019-NCOV) in China Estimating the value of political connections The mortality cost of political connections Subnational debt of china: the politics-finance nexus Economic fluctuations and growth Credit spreads and business cycle fluctuations Do politically connected boards affect firm value? The impact of government intervention on corporate investment allocations and efficiency: evidence from China A new capital regulation for large financial institutions Political connections and the cost of bank loans The risk of implicit guarantees: evidence from the shadow interbank market in China Local crowding out in China Hayek, local information, and commanding heights: Decentralizing state-owned enterprises in China Population flow drives spatio-temporal distribution of COVID-19 in China Capital market, financial institutions and corporate finance in China The value and real effects of implicit government guarantees Let's take the con out of econometrics Political connections, financing and firm performance: evidence from Chinese private firms Policy burdens, accountability, and the soft budget constraint Implicit government guarantee and the pricing of Chinese LGFV debt What promotes/prevents firm bond issuance in emerging economies: Bank-firm relationship or information asymmetry? Inventory Capacity and Corporate Bond Offerings Impacts of social and economic factors on the transmission of coronavirus disease 2019 (COVID-19) in China Government credit, a double-edged sword: evidence from the China development bank How to explain corporate investment heterogeneity in China's new normal: Structural models with state-owned property rights Credit rationing in markets with imperfect information Mitigating political uncertainty Is Trust Priced? Evidence from the Bond Market Does mixed-ownership reform improve SOE's innovation? Evidence from China's state ownership EBs are regulated by the National Development and Reform Commission (NDRC), a powerful government agency overseeing SOEs. ETCBs are issued in the exchange market and regulated by China Securities Regulatory Commission (CSRC) Yes