Does cross-industry loan diversification reduce bank systemic risk? Evidence from listed banks in China

ABSTRACT Banking plays a crucial role as the most significant financial intermediary, and the effectiveness of its diversification strategy aimed at reducing individual bank systemic risk is a significant topic in both theory and practice. In this study, we investigate the impact of cross-industry loan diversification on bank systemic risk using data from listed banks, and reveal a significant positive correlation, indicating a clear‘diversification systemic risk’ effect. This effect remains robust across a series of robustness tests. Further research reveals the influence of cross-industry loan diversification on bank systemic risk operates primarily through the channel of interbank asset similarity. Additionally, we find the‘diversification systemic risk’ effect is particularly prominent among joint-stock banks, those with high interbank deposits，and those with high real estate loans. These findings hold important reference value for the ongoing reforms and practices in China aimed at preventing bank systemic risk.


Introduction
Given the importance of commercial banks in the economy, bank risk and bank risk management have always been topics of concern for regulatory authorities and scholars (Chu et al., 2020;Fang, 2016;Fang & Zheng, 2016;Guo & Zhao, 2017;Wang & Li, 2021).
The financial crisis triggered by the bankruptcy of systemically important commercial banks in 2008 has raised the anxiety of governments and international organisations regarding the current state of bank risk management, leading to major transformations in bank regulation and business models (Yang et al., 2020;Zhang, 2010).Specifically in China, with the economic growth downturn and deepening reforms, the Chinese government has shown unprecedented levels of concern for bank risk and its management.The report of the 20th National Congress of the Communist Party of China proposed 'deepen financial system reforms, enhance the capability of the financial sector to serve the real economy', and 'strengthen the financial stability guarantee system, while guarding against systemic financial risks'.In July 2017, during the National Financial Work Conference in China, it was emphasised that preventing and defusing financial risks is a perpetual theme of financial work and should be given greater importance.As a result, a variety of measures aimed at reducing bank risks have been introduced.In terms of bank supervision, efforts have been made to strengthen prudential supervision of banks' macro management, enhance capital constraints on the banking industry based on Basel III agreements, and promote reforms in bank business models.Regulatory authorities have advocated diversification of bank operations, which has been actively implemented by numerous banking institutions.
Cross-industry loan diversification is a major form of diversified operation for commercial banks (Zhang, 2004).Bank diversification refers to the expansion in multiple products and services, including derivative new business, seeking new markets and new customers and maintain competitive advantages (Stiroh & Rumble, 2006).Generally, bank diversification usually includes cross-industry loan diversification, geographic diversification and income diversification (Wang & Kang, 2022;Zhang, 2004).Specifically, cross-industry loan diversification refers to the diversification of the layout of the loan business.Banks previously mainly engaged in industrial and commercial enterprise may enter the real estate mortgage loan market, forming the distribution of loans in kinds of industries.Geographical diversification refers to the expansion of bank business in diversified geography.Income diversification refers to the optimal combination between interest-bearing and noninterest-bearing business (Wang & Tian, 2014).
In recent years, China's national economy has achieved a series of remarkable accomplishments, largely due to the development and diversification of various industries.The development of industries, especially key sectors of the national economy, relies on the strong credit support provided by the banking sector through cross-industry loan diversification (Fan et al., 2019).However, due to the scarcity of credit resources, during the early stages of economic development, bank loans tend to concentrate in low-risk or traditional industries, leading to an imbalance where some high-risk and emerging industries face difficulties in obtaining loans.To mitigate the risks associated with concentrated bank lending, the government has continuously introduced institutional measures to encourage banks to expand the breadth of their loan portfolios, thus enhancing the level of cross-industry loan diversification (Li & Zhu, 2021).
In 2004, the National Development and Reform Commission, the People's Bank of China and the former Banking Regulatory Commission stipulated that, banks should focus on industry development, implement national industrial policies, and adjust credit allocation promptly in certain industries in accordance with the 'Catalog for Restricting Lowlevel Repetitive Construction in Certain Industries' within the specified range by raising loan interest rates and reclaiming loan approval authority. 1This policy clearly guided the banking industry to expand the scope of loan allocation to industries.
In 2021, the People's Bank of China further directed financial institutions to support the development of green emerging industries and evaluate green finance and other loan businesses based on both quantitative and qualitative indicators, guiding banks to actively channel loans to key emerging sectors. 2t can be said that bank loans have been strategically deployed at the industry level in the national economy under policy guidance.On the other hand, regulatory authorities have also issued policies to limit the proportion of loans concentrated in certain industries, making adjustments and regulations in sectors with high loan-to-asset ratios, such as real estate.Specifically, in 2021, the People's Bank of China and the former China Banking and Insurance Regulatory Commission issued a joint notice stipulating that the proportion of real estate loans and personal housing loans in banking institutions should not exceed the upper limits determined by the People's Bank of China and the China Banking and Insurance Regulatory Commission for real estate loan ratios and personal housing loan ratios.This regulation aims to strictly control the phenomenon of large-scale concentrated loans. 3egarding industry restrictions and prohibitions, the state promulgated the first 'Catalogue for Guiding Industrial Restructuring' in 2005, which effectively guided credit allocation to industries.Subsequently, the content of the catalogue has been adjusted several times based on the actual situation, and the latest version is the 'Catalogue for Guiding Industrial Restructuring (2021)'.The catalogue includes 190 items in the restricted category, such as 'oil-fired boilers and power generators primarily for power generation', and 399 items in the phased-out category, such as 'outdated production process equipment'.Based on this guidance, major banks also publish their own 'Guidelines for Industry Credit Allocation' each year, which primarily focus on industries with overcapacity and specify the industries that banks are prohibited from entering.
Under the backdrop of international bank diversification strategies and the urgent need for their own development, the banking industry in China has embarked on a path of diversification in recent years.As shown in Figure 1, the annual trend of cross-industry loan diversification of listed banks in China has been continuously increasing from 2007 to 2019.Commercial banks, by leveraging internal resources and expanding their loan services to various industries, have become an extremely important trend in the development of the banking sector in China (Liu et al., 2012;Wang & Tian, 2014).
Under the strong advocacy of regulatory authorities in China, the cross-industry lending in the banking industry has experienced rapid development, providing robust support for the 'fairness and efficiency' development of various industries.However, it is important not to overlook the potential risks associated with the rapid expansion and increasing diversification of loans across industries, as well as the growing similarity in the industry structure of bank lending.Banks play a vital intermediation role in the financial system, and bank risks leading to banking crises can have severe consequences.Effectively mitigating the risks, particularly systemic risks, in the rapid diversification process of banks has been a focus of attention for regulatory authorities and academia, both domestically and internationally (Chu et al., 2020;Fang, 2016;Fang & Zheng, 2016;Guo & Zhao, 2017;Wang & Li, 2021).
Due to the different stages of economic development in various industries and regions, as well as the different driving factors behind industry and regional development, investing in different industries and regions can reduce the risk of the asset portfolio compared to investing in a single target (Markowitz, 1952;Sharp, 1964).Viewing cross-industry loans by banks as an investment portfolio, according to the portfolio theory, diversification across industries can reduce bank risks (Deng & Elyasiani, 2008;Goetz et al., 2016;Hughes et al., 1999).However, the assumptions of portfolio theory are relatively stringent, and some scholars have questioned its applicability to loan diversification.Scholars holding this view argue that the theoretical foundation of portfolio theory, which assumes that investors base their investment choices on expected returns and the variance of returns, does not effectively apply to loan diversification (Yu, 2015).They contend that the return distribution of loan portfolios in banks is highly asymmetric, making variance an inadequate measure of risk, and the concepts of expected returns and variance cannot solely serve as the basis for this theory (Wagner, 2008(Wagner, , 2010;;Wei, 2010).
At the same time, portfolio diversification reduces diversification risks, which are the non-systematic risk components of individual banks.The reduction of non-systematic risks for individual banks does not necessarily imply a reduction in the overall risk of individual banks, let alone a reduction in the overall risk of the banking industry.The systemic risk in the banking industry is not simply the sum of individual risks of bank institutions and the inherent characteristics of risk accumulation and contagion between banks in the banking industry mean that the reduction of individual bank risks does not necessarily lead to a reduction in the overall systemic risk of the banking system (Yao et al., 2019).From a holistic perspective, the likelihood and amount of similar assets held by different banks increase with the diversification of bank loans.The increase in asset similarity between banks increases the possibility of simultaneous failures of banks and indirect contagion caused by price losses, thereby significantly increasing systemic risk in the banking sector (Chu et al., 2020;Duarte & Eisenbach, 2015;Fang & Zheng, 2016;Wagner, 2010Wagner, , 2011)).Furthermore, the increase in similarity of assets between banks may lead to overcapacity in different industries and cause fluctuations in the real economy (Liu & Cai, 2016).In general, we hold that the reduction of idiosyncratic risks for individual banks may not necessarily represent a weakening of systemic risk.Whether the advocated diversification reforms, actively practiced by many banks, can truly reduce the level of systemic risk in the banking industry remains an empirical question that requires further investigation.
Therefore, we construct a sample of 33 listed banks in China from 2007 to 2019, adopt OLS and panel fixed effect research methods, and try to study the relationship between cross-industry loan diversification in China and bank systemic risk.We find that as the degree of cross-industry diversification increases, the level of systemic risk in banks does not decrease.On the contrary, cross-industry loan diversification exacerbates systemic risk, indicating the presence of a 'diversification systemic risk' effect.The conclusion is valid with a series of robustness tests.
We conduct further tests to explore how cross-industry loan diversification worsens systemic risk.The further tests show that the 'asset similarity' among banks is the conceivable channel through which cross-industry loan diversification worsens systemic risk.Asset similarity refers to the similarity of assets held by banks and other banks (Allen et al., 2012).The higher the similarity of inter-bank assets, the greater level of risk they jointly exposed to, resulting in higher systemic risk.In this way, the increased asset similarity among banks due to cross-industry diversification is an important channel that exacerbates systemic risk in banks (Chu et al., 2020;Fang & Zheng, 2016;Wang et al., 2022).Additionally, we examine the cross-sectional differences in 'diversification systemic risk' effect and find out the impact of cross-industry diversification on systemic risk varies, particularly among joint-stock banks, banks with high inter-bank deposits and high real estate loans.This article makes three major contributions.Firstly previous studies analysed the risk consequences of bank diversification using portfolio theory, confirming that diversified portfolio of loans can reduce risk.This article, based on the limited applicability of portfolio theory and the formation mechanism of systemic risk (Chu et al., 2020;Wagner, 2008), arrives at the conclusion that cross-industry loan diversification cannot effectively reduce the overall bank risk and may even worsen systemic risk.This enriches the theoretical basis and empirical evidence regarding the risk consequence of bank loan diversification, demonstrating a certain level of innovation.
Secondly, for a long time, research on the factors of systemic risk has always been a major concern and a crucial academic topic, given the potential severe consequences of systemic risk.Existing literature usually examine the issue from internal and external factors.Concerning internal factors of systemic risk, previous research has focused on the impact of bank runs on liquidity (Diamond & Dybvig, 1983), bank risk management (Dani et al., 2004), maturity mismatch in loans and deposits (Aikman et al., 2017;Brunnermeier & Oehmke, 2013), and interbank network diffusion and contagion (Allen & Gale, 2000, Fang, 2016;Gong et al., 2020) on systemic risk.In this study, cross-industry loan diversification is regarded as a new internal factor, complementing the relevant literature research on the influencing factors of systemic risk.
Lastly, there is relatively limited literature in China that examines the factors contributing to systemic financial risk from a micro perspective.Previous studies have mostly focused on studying the causes and consequences of systemic risk from a macro perspective (Fang & Zheng, 2016).This study uses micro-level industry loan and bank financial data, providing empirical evidence for the occurrence of increased systemic risk in commercial banks caused by cross-industry loan diversification, and offering important decision-making references for regulatory policy formulation.
The remainder of the paper proceeds as follows: The next section is literature review and raises our points of study.Section 3 introduces the sample selection process, and the design of the regression model.Section 4 and 5 report the empirical results.Section 6 offers our conclusions.

Definition and characteristics of systemic risk
Before measuring systematic risk, it is necessary to accurately define systematic risk.Currently, there is no universally accepted definition of systemic risk, which itself indicates that systemic risk is a complex issue, requiring further exploration for this research question.Specifically, the definition of systemic risk is mainly approached from three dimensions.
Firstly, from the perspective of risk contagion, systemic risk refers to the risk of endogenous events, such as the collapse of financial institutions or market crashes, spreading from one institution or market to multiple institutions and markets, resulting in the continuous diffusion of losses within the financial system (Hart & Zingales, 2009;Kaufman & Scott, 2003).
Secondly, from the perspective of the scope of harm, Bernake proposed in 2009 that systemic risk refers to the event that can threaten the entire financial system and the macro economy.In the same year, the European Central Bank defined systemic risk as the risk of extreme fragility in the financial system, widespread financial instability, and difficulties in the functioning of the financial system.
Lastly, from the perspective of impact on the real economy, the G20 finance ministers and central bank governors' report in 2010 defined systemic risk as the risk of impaired or disrupted financial service processes that could cause severe adverse effects on the real economy.
The definition of systemic risk is related to the concepts of financial crisis and financial fragility but has essential differences.The definition of a financial crisis is a sharp, temporary, and super-cyclical deterioration in all or most financial indicators, such as shortterm interest rates, asset prices (including securities, real estate, and land assets), the number of business bankruptcies, and financial institution failures.A financial crisis is a binary variable with a value indicating whether it happens or not.It is the result of the accumulation of systemic risk, representing a specific stage and state of systemic risk.
Another similar concept to systemic risk is financial fragility.Generally speaking, financial fragility refers to the inherent attribute of a financial system that is more prone to failure due to the characteristics of highly leveraged operations.It represents a financially vulnerable state (Huang, 2001;Minsky, 1992).Financial fragility theory mainly studies the inherent defects of financial system from the institutional level, which is one of the factors of systemic risk.Systemic risk, on the other hand, is an observed or probabilistic phenomenon, while financial fragility is one of the causes that lead to or amplify this probability (Zhang, 2010).
While there is no unified definition for systemic risk, the definitions from different dimensions share the following commonalities.Firstly, systemic nature: systemic risk goes beyond the individual risks of single financial institutions and focuses on the risks of the entire financial system or significant components of it.It exhibits interdependence and systemic characteristics, surpassing the simple aggregation of individual risks.Secondly, externalities: systemic risk involves externalities, meaning that significant losses in a single financial or market can trigger a chain reaction affecting other financial institutions or markets.Systemic risk is collectively borne by all participants within the financial system, implying negative externalities.Thirdly, contagion effect: systemic risk can lead to the transmission of risk from the financial system to the real economy, causing significant disruptions and damage to the real economy.
Given the aforementioned definitions and characteristics of systemic risk, we define systemic risk for banks as the risk that can potentially disrupt the service functions of banking institutions, increase uncertainty in the financial system, and pose severe harm to the real economy.

Determinants of systemic risk
The occurrence of financial crises can cause severe damage to the real economy and social stability, as seen in the global financial crisis that erupted in 2008, which has not been fully eliminated in certain industries and companies to this day.As financial crises are the result and manifestation of the accumulation of systemic risk, representing a specific state of systemic risk, research on systemic risk has garnered significant attention from scholars and regulatory authorities worldwide (Wang & Li, 2021).A review of existing domestic and international literature reveals that the study of systemic risk causes typically approaches from the following perspectives: external factors such as macroeconomic shocks and market volatility, and internal factors within financial institutions.
External factors are risk factors of macroeconomic shocks and market volatility, such as macroeconomic downturn (De Bandt & Hartmann, 2000), herding effect in financial markets (Acharya, 2009) and moral hazard associated with financial safety nets (Ji et al., 2018).In addition, scholars have also studied the mechanism of implementing different monetary policies on systemic risks (Kupiec & Nickerson, 2004).
Internal factors primarily focus on individual financial institutions or the entire system, extracting risk factors from these institutions to examine the causes of systemic risk.One important theoretical foundation within the realm of internal factors is the Diamond-Dybvig (D-D) model, proposed by Diamond and Dybvig (1983).The D-D model reflects the liquidity problems caused by bank runs, which can lead to systemic risk and banking crises.Dani et al. (2004) suggests that risk management methods employed by banks can contribute to procyclicality and increase systemic risk.Furthermore, factors such as maturity mismatch in bank loans and deposits proposed by Brunnermeier and Oehmke (2013) and Aikman et al. (2017), distorted incentive mechanisms in individual financial institutions as suggested by Schwarcz and Anabtawi (2011), are all significant contributors to increased systemic risk in the banking sector.
In addition, scholars have found that risks can spread and contagion can occur through risk spillover among financial institutions and interconnected networks.This is another important factor exacerbating systemic risk and triggering crises (e.g. the 2008 financial crisis).In the literature on interbank networks, early contributions applied network analysis methods, constructing complex network models and indicators to study systemic risk (Allen & Gale, 2000).Subsequently, scholars have developed more analytical methods in this field.For example, Gong et al., 2020 study the contagion of systemic risk within the financial system by establishing an information spillover network among listed financial institutions in China using variance decomposition network methods.They identify systemically important financial institutions in financial network risk contagion.Another widely studied issue is the channels of systemic risk contagion.For example, Fang (2016) constructs an asset-liability interconnected network model incorporating bank bankruptcy mechanisms and deleveraging mechanisms to investigate systemic risk contagion channels.The study finds that the most significant channels in a crisis are deleveraging and bank debt default channels.
In terms of the causes of systemic risk in China, due to the relatively healthy operation of the economy and its large scale, with a positive economic growth outlook, external factors are unlikely to directly generate significant fluctuations in bank operations or a high probability of bank crises.Therefore, the largest source of systemic risk is likely to come from internal factors within financial institutions.

Specification of systemic risk
With the deepening study on definition and determinants of systemic risk, the specification of systemic risk have continuously evolved.As illustrated by a number of academic studies, specification indicators are also mainly categorised into two groups: trait-based indicators that focus on the internal factors of a specific type of systemic risk, and global risk measurement indicators, mainly using network analysis methods to measure contagion and interconnections among financial institutions.Specifically, the first category is used to describe the bank liquidity crisis (Brunnermeier & Oehmke, 2013;Brunnermeier & Pedersen, 2009;Shleifer & Vishny, 1992), and systemic risks in the shadow banking sector (Li & Xue, 2014).The second category mainly includes network analysis (Allen & Gale, 2000;Diebold & Yilmaz, 2014;Glasserman & Young, 2015;Yang et al., 2020), Contingent Claims Analysis method (CCA) or the structured model method (Fan et al., 2013(Fan et al., , 2018;;Gray & Jobst, 2010;Vassalou & Xing, 2014) and the inter-bank asset-liability network model (Duffie & Zhu, 2011, Fang, 2016;Fang & Huang, 2019).
Afterwards, the specification is no longer limited to the above traditional network analysis method, and scholars have put forward simplified analysis methods.These methods focus on the tail characteristics of the asset returns of financial institutions and have been well developed after the financial crisis.For example, 'value-at-risk (VaR)' is used to measure the maximum possible loss of financial assets within a certain holding period and given confidence level.The 'expected shortfall' (ES) method proposed by Diebold and Yilmaz (2014) constructs the correlation matrix through variance decomposition.
While the VaR and ES methods have relatively simple calculation processes and certain generalisability, their shortcomings include the potential underestimation of risk spillover effects among different financial institutions.Adrian and Brunnermeier (2016) introduce the 'Delta conditional value-at-risk' (ΔCoVaR) method, which improves upon the VaR method by using the difference in conditional value-at-risk to measure systemic risk of financial institutions.This method examines the impact of a specific financial institution on the entire financial system during a crisis by analysing the difference between the conditional value-at-risk when facing financial distress and the normal operational valueat-risk.It effectively measures the systemic financial risk contribution of the institution and serves as a standard for determining its required capital to prevent risk.The ΔCoVaR method is currently the most recent and effective marginal contribution method for measuring financial risk within the simplified analysis framework and has gained widespread recognition.Domestic scholars have conducted empirical analyses on systemic risk based on the ΔCoVaR method.For example, Guo and Zhao (2017) use the ΔCoVaR method to examine the relationship and mechanism between bank deposit competition and systemic risk.
At present, China's financial sector is still in a risk-prone period.Under the pressure of multiple factors, macro-prudential supervision requires close attention to the crossmarket and cross-industry contagion of financial risks.The simplified analysis approach treats the entire market as a portfolio of financial institutions and analyzes the risk contribution levels of institutions based on financial market data.It focuses on the synergy and feedback effects between institutions and the financial system, making it a suitable research method for the financial environment and regulatory requirements in China.

Cross-industry loan diversification and bank systemic risk
Bank risk and bank risk management have always been topics of common concern for regulators and scholars (Chu et al., 2020;Fang, 2016;Fang & Zheng, 2016;Guo & Zhao, 2017).Since 2008, bank risk and its management have attracted high attention from governments and international organisations, leading to significant changes in bank regulation and business models worldwide (Wang & Tian, 2014;Zhang, 2010).In addition, macro-level strengthening of prudent supervision of bank macro-management and capital constraints on the bank industry through the Basel III Accord, reform measures regarding bank business models have also emerged.Bank diversification has been advocated by the regulatory authorities and actively implemented by many banking institutions.The analysis of China's systemic risk from the perspective of bank diversification is one of the important academic topics in recent years (Chu et al., 2020).
As a financial intermediary in China's capital market, banks' loan resources, which are continuously allocated to the economy and society, are not only one of the most basic bank assets but also the most important destination for bank fund investments.In early research on bank diversification and bank risk, traditional strategic theory and portfolio theory are the main research foundations.Scholars believe that funds invested in different sectors and industries can be considered as a form of portfolio.Due to the 'coinsurance effect' of diversification (Boot & Schmeits, 2000;Lewellen, 1971) and the 'market portfolio theory' (Markowitz, 1952;Sharp, 1964), funds invested across different industries and regions with different stages of economic development, different driving factors for industry and regional development will reduce the risk of cross-industry and cross-regional investment portfolios (Boot & Schmeits, 2000;Lewellen, 1971;Markowitz, 1952;Sharp, 1964).
When banks adopt the advocated diversification strategy by regulatory authorities, from the perspective of individual banks, loan diversification across industries enables the investment portfolio of bank assets to shift from a single focus on certain pillar industries to serving and covering more industries, enriching the bank's loan products, and diversifying the bank's investment portfolio.When a particular industry is hit by an external shock and a bank's client in that industry experiences credit delinquency or even default, loan diversification across industries can reduce bank losses, smooth the bank's operating profits, and reduce the risk of bank loans (Deng & Elyasiani, 2008;Goetz et al., 2016;Hughes et al., 1999).Liu et al. (2012) use the standard deviation of bank asset returns as a proxy variable for individual bank non-systemic risk and the Herfindahl index of bank revenue structure as a proxy variable for the degree of diversification, concluding that diversification by Chinese banks can reduce their individual risk.Wang and Li (2021) use financial data of Chinese city commercial banks and the distribution data of their branches, using the non-performing loan ratio as a proxy variable for bank risk.They find that geographic diversification of Chinese city commercial banks can significantly reduce their individual risk once financial risk exceeds a certain threshold level.
However, diversification increases organisational complexity and makes it difficult to effectively regulate and address operational issues, and it may also lead to higher individual risk (Berger et al., 2005;Demsetz & Strahan, 1997;Winton, 1999).Zhou and Li (2011) find no significant relationship between the diversification of commercial bank income structure and individual bank risk in China.The reduction in individual bank risk is mainly attributed to the decrease in interest income volatility, while an increase in the proportion of non-interest income actually increases the volatility of non-interest income and contributes to the overall risk.
Given the inconsistent conclusions from domestic and international studies, some literature provides explanations for the underlying reasons for these discrepancies.On the one hand, some scholars argue that portfolio theory may not be suitable as the theoretical foundation for bank loan diversification.The assumptions of portfolio theory are based on investors' expectations of return and the variance of return as the basis for investment selection.Investors balance expected returns and risks, but in bank loan portfolios, whether diversified across industries or geographically, the distribution of returns is highly asymmetric (Yu, 2015).Therefore, variance is no longer suitable for measuring risk in bank loan portfolios, and the concepts of expected return and variance cannot serve as the basis for analysing loan diversification and bank risk issues (Wagner, 2008(Wagner, , 2010)).Some studies have found that loan diversification may have a negative effect on risk.A study of 42 countries finds that financial conglomerates engaged in various types of lending have a market value lower than the sum of the market values of specialised lenders, indicating a 'diversification discount'.The study concludes that loan diversification is associated with higher risk or lower risk-adjusted performance.Chinese scholars have conducted research on this issue as well.Wei (2010) argued that bank loan diversification does not meet the theoretical prerequisites of portfolio theory and, therefore, does not necessarily reduce the risk of commercial banks.
On the other hand, systemic risk, which is a key concern and focus of regulatory authorities, is not equivalent to a simple aggregation of individual bank risks (Yao et al., 2019).It is important to distinguish between the mechanisms of systemic risk and those of individual bank risk.Some scholars have enriched the research on the consequences of bank diversification risk from the perspective of internal factors influencing systemic risk.
First, existing literature suggests that systemic risk in banks refers to the risk that can force the interruption of their banking services, increase uncertainty in the financial system, and cause serious harm to the real economy.Regarding systemic risk, there is no direct relationship between bank systemic risk and non-systemic risk (diversifiable risk) at the individual bank level (Yao et al., 2019).Systemic risk is inherent in banking and other financial markets.While some studies suggest that loan diversification can reduce the non-systemic risk of individual banks, it cannot disperse or eliminate individual systemic risk through diversification.Therefore, it does not mean that overall bank risk or systemic risk naturally decreases.
Second, scholars have conducted research on systemic risk from the perspective of the overall asset distribution in the banking industry.Studies suggest that cross-industry loan diversification by banks leads to an increase in the variety of bank assets, which may be considered an optimal choice for risk reduction for an individual bank.However, when considering the banking industry as a whole, the asset structures held by banks become increasingly similar with the level of diversification increasing.This increases the risk of multiple banks failing simultaneously, thereby raising the likelihood of systemic risk and banking crises.Wagner (2010) explores the relationship between bank diversification and systemic risk using a theoretical model involving two banks.The study concludes that limited diversification is feasible, while extensive and complete diversification is inefficient.It demonstrates that when banks are completely non-diversified, a shock affecting one bank and causing its assets to rapidly depreciate below its liabilities would lead to the bankruptcy alone, without causing a crisis.However, when banks are completely diversified, holding identical asset portfolios, any extreme external shock would lead to the simultaneous collapse of both banks, resulting in a banking crisis that surpasses the failure of a single bank.In other words, diversification increases the possibility of multiple banks failing simultaneously, thereby raising systemic risk and the likelihood of a banking crisis.Ibragimov et al. (2011) argue that when multiple banks hold identical assets, a shock can simultaneously disrupt all institutions, resulting in costly consequences for society.Compared to the bankruptcy of a single bank, the recovery time required for the financial system and the economy will be longer after all institutions suffer certain losses.The loss of the intermediation function of banks and the slow recovery time incur significant and sustained social costs (Bernake, 1983).
Furthermore, some literature suggests that diversification cannot truly prevent systemic crises.When the value of each bank's assets is lower than its liabilities, regardless of the form of asset composition, systemic crises cannot be truly resolved through diversification (Wagner, 2008).In conclusion, the aforementioned studies indicate that the mechanism of systemic risk differs from that of individual bank risk factors.Crossindustry loan diversification among banks exacerbates systemic risk beyond the optimal level of diversification, and diversification behaviour cannot truly avoid systemic crises.
Third, the above research analyzes the relationship between diversification and systemic risk from the perspective of the internal formation mechanism of systemic risk.However, the existing models did not consider the interconnectivity between bank institutions.Subsequently, scholars explored the influence of interbank markets, interest rate levels, and the potential contagion paths of runs on banks on systemic risk, thus expanding the theoretical extension of the basic model constructed by Wagner (2010) involving two banks.Fang and Zheng (2016) depict the contagion of systemic risk between risk-generating banks and risk-bearing banks from a micro-institutional perspective.Risk-generating banks are often systemically important banks that suffer significant exogenous shocks and have high contagion potential, while risk-bearing banks are usually systemically important banks that suffer relatively smaller exogenous shocks and have asset structures similar to risk-generating banks.The similarity of asset structures between banks exacerbates the contagion of systemic risk.Chu et al. (2020) study the issue of cross-regional loan diversification and systemic risk using data from the United States.The similar asset structures resulting from geographic expansion among banking institutions increased the possibility of multiple banks failing simultaneously, further leading to increased pressures on asset price depreciation, thus generating negative externalities of systemic risk.
Finally, some scholars have studied the issue of diversification systemic risk from the perspective of external shock heterogeneity.Ibragimov et al. (2011) constructed a theoretical model on the relationship between financial institution diversification and systemic risk.The research model suggests that under extreme 'fat-tail risk' shocks, diversification among financial institutions leads to risk sharing among them, which can exacerbate the consequences of systemic risk.However, when the probabilities of both extremes are low, systemic risk does not worsen.
The aforementioned studies on the risk of diversified bank loans have important theoretical and practical implications for regulatory agencies implementing industrial policies and macroprudential regulation.However, most of the previous research has focused on individual bank risk, with limited empirical research on the impact of cross-industry loan diversification on systemic risk.Existing literature mainly relies on theoretical models and case studies, and empirical studies on the internal impact mechanism of cross-industry loan diversification as a factor influencing systemic risk are relatively scarce.Chu et al. (2020) conducted research on the systemic risk issue of cross-regional loan diversification in the United States and examined the effects of geographic expansion and the resulting risk consequences.Due to significant differences in the institutional backgrounds of the banking industries in China and the United States, the gradual relaxation of geographic regulatory systems in the United States since 2007 provides an opportunity to investigate the impact of geographical expansion of banking business on systemic risk under exogenous regulatory shocks.It also provides a research direction for examining cross-industry loan diversification.While the number and distribution density of bank branch expansions in China still differ significantly from those in foreign countries, China has made significant progress in terms of loan industry diversification.Therefore, this study attempts to investigate the relationship between cross-industry loan diversification and systemic risk, aiming to complement the study of systemic risk determinants.
Some domestic literature has attempted to investigate the impact of loan diversification on systemic risk.For instance, Zhang (2004) analyzes the current situation of cross-industry loan diversification in China, but it lacks theoretical and mechanism analysis.Yu (2015) studies the systemic risk consequences of crossindustry loan diversification, but it lacks a theoretical basis and empirical proof.Wang and Li (2021) study the relationship between loan diversification and bank risk using the financial data of Chinese city commercial banks and the data of their branches, but it does not address the consequences of systemic risk and corresponding mechanistic analysis.Wang et al. ( 2022) make a preliminary attempt on the channel of asset similarity mechanism, and study the relationship between diversification and bank systemic risk using the bank systemic loss model.They focus on bank investment diversification, specifically measuring diversification by the amount of investment in different types of assets.The topic of this paper is the systemic risk caused by the similarity of small and medium-sized commercial banks.
In contrast, unlike Wang et al. (2022), we focus on large listed banks within the Chinese system and examine the cross-industry loans of these banks and their significant role in systemic risk in the banking sector.For research theme of cross-industry bank loans complements the research content of diversified bank investment by Wang et al. (2022), contributing to a more comprehensive and profound understanding of the relationship between cross-industry diversification of bank loans and systemic risk in the context of China's unique system.
When examining the relationship between cross-industry loan diversification and systemic risk, we consider the heterogeneity of banks in terms of types, inter-bank deposits, and real estate loans, and use cross-sectional differences affected by such heterogeneity to make inter-group comparisons.In contrast, Wang et al. (2022) only tested systemic risks.To more comprehensively measure the risk consequences of crossindustry loan diversification on banks, we further tested the impact of cross-industry loan diversification on the non-systemic risks and overall risks of individual banks.This expanded study also helps to establish a better causal relationship between loan crossindustry diversification and systemic risk.
For the aforementioned reasons, we use the semi-annual data from 2007 to 2019 of 33 listed banks in China as a sample to empirically test the relationship between crossindustry loan diversification and bank systemic risk.Our aim is to assess whether the cross-industry loan diversification advocated by regulatory authorities and actively practiced by many banks can effectively reduce the systemic risk level of the banking industry in the emerging market scenario of China.

Sample selection and data sources
In order to accurately measure bank systemic risk, the financial data of banks is obtained from the CSMAR database, while bank stock daily returns and other stock market data are from the Wind database.To construct a bank systemic risk indicator using daily or weekly stock returns, we select listed banks in China as the sample.Since bank industry diversification is only disclosed in banks' semi-annual and annual financial reports, we use data from the period of 2007 to 2019, including semi-annual and annual data.We remove any missing financial and market data from the sample, resulting a total of 438 bank-semiannual samples of 33 listed banks, including 108,766 data points for the construction of bank systemic risk indicator and 7,359 data points for the construction of cross-industry loan indicator.Lately, to eliminate the impact of extreme values, we winsorise all continuous variables by 1%.

Research model
Regarding the research model, we follow the approach of Chu et al. (2020) and construct the following model to examine the relationship between cross-industry loan diversification and systemic risk: The regression analysis is carried out using the ordinary least square method and nonequilibrium panel fixed-effect model, while controlling for bank-semi-annual fixed effect.The explained variable 'bank systemic risk' (CoVaR).To measure the systemic risk of Chinese banks, we adopt the measurement method of 'Delta conditional value-at-risk' (ΔCoVaR) proposed by Adrian and Brunnermeier (2016).ΔCoVaR refers to the risk value of the entire banking system when a particular bank is in a certain state.The contribution of an individual bank to the banking system is measured by the difference between the risk value of the banking system in a crisis state and in a normal state, known as ΔCoVaR.In this way, ΔCoVaR can be used to measure the contribution of an individual bank to systemic risk.The estimation process is as follows: First, to construct quantile regression equation: Where, R i,t represents the daily return rate of the bank, calculated using the daily closing price data of the listed bank, R t = 100×ln(P t /P t-1 ),where is the closing price of the listed bank; R system,t is the weighted average return of the daily return rate of all listed banks under the market model, weighted by the market value of each bank; M represents a series of state variables.Referring to literature (Adrian & Brunnermeier, 2016;Bai & Shi, 2014;Guo & Zhao, 2017), we select market returns (represented by the volatility of the CSI 300 Index as the proxy for average market returns), short-term liquidity trend variable (representing the degree of short-term financial market liquidity tightening, represented by the difference between the 3-month bank pledged repo rate and the 3-month government bond rate), interest rate spread trend variables (represented by the change in the 3-month government bond rate), changes in the slope of the Chinese yield curve (representing changes in the Chinese economic cycle, represented by the yield spread between the 10-year and 3-month Chinese government bonds), and changes in the slope of the US yield curve (representing changes in the global economic cycle, represented by the yield spread between the 10-year and 3-month US Treasury bonds).All of the state variables are lagged by one period to eliminate endogeneity and contemporaneous effects.
Next, using quantile regression, the corresponding coefficient estimates are obtained and plugged into Equation (3) and Equation (4), resulting in the calculation of a bank's value at risk and the conditional value at risk of the banking system.
Finally, the marginal contribution of Bank i to systemic risk, ΔCoVaR, can be obtained by calculating the difference between the conditional value at risk of the banking system when Bank i is in a crisis state (q = 0.05) and when it is in a normal state (q = 0.5).The formula for ΔCoVaR is as follows: The calculated values are generally negative.To obtain the semi-annual measure of systemic risk for each bank, take the absolute value of ΔCoVaR and calculate its mean.A larger value of this measure indicates greater systemic risk for the bank.
The explanatory variable in the model is the bank loan cross-industry diversification (1-HHI).The Index is derived from the Herfindahl-Hirschman Index (HHI), which is calculated based on the number of loans issued by listed banks to different industries every six months.The HHI index is commonly used to measure the concentration of a certain characteristic within a firm, and in this study, it is used to measure the concentration of cross-industry loan.The HHI index is calculated as follows: The HHI index is calculated based on the number of loans allocated to each industry by a bank in each half-year.X i represents the number of loans allocated to industry i by bank i, X represents the total number of loans allocated by the bank, X i/ X represents the proportion of loans allocated to industry i by the commercial bank.N is the total number of industries to which the bank allocates loans. 5A higher HHI value indicates a more concentrated allocation of loans meaning that the bank's loans are focused on specific industries.On the other hand, the cross-industry bank loan diversification variable is 1-HHI, with a value between 0-1.The closer it is to 1 indicates that the bank's loans are more diversified across different industries.
According to existing literature, we control the following variables in the model: Bank size (Size and Size 2 ), measured by the natural logarithm of total assets.The square of the natural logarithm of total assets is included to capture the nonlinear effect; Funding structure (ST_funding), represents short-term non-deposit funding, including interbank borrowing, deposit certificates, and short-term bonds.It is calculated as the ratio of nondeposit short-term funding to the sum of deposits and non-deposit short-term funding, multiplied by 100%; Bank profitability (Roa), measured by the return on assets (ROA), multiplied by 100%; Book-to-market ratio (Book_to_market), represents the ratio of equity book value to market value, multiplied by 100%; Non-interest income (Noninterest), calculated as the ratio of non-interest income to total operating income, multiplied by 100%; Loan loss provisions (Provision), measured as the ratio of loan loss provision to total loans; Diversity of income (Income_diversity), calculated as 1-|(net interest income -noninterest income)/bankingoperating income|; Diversity of asset (Asset_diversity), calculated as 1-|(net loans -other interest-bearing assets)/total interest-bearing assets|.Other interest-bearing assets include cash and deposits with central banks, net interbank placements, and net buybacks of financial assets; Personal loan (Personal_loan_ratio), calculated as the ratio of the personal loan amount to the total bank loan, multiplied by 100%, to capture the impact of personal loans on systemic risk; Geographic diversification (Geo_diversity), represents the geographic expansion of the bank's branch network.According to Wang et al. (2012), the Herfindahl-Hirschmann Index is calculated based on the number of branches that a bank has in different cities. 6 Geo_diversity is then calculated as 1-HHI G ,with a value closer to 1 indicating a higher level of geographic diversification in the bank's branch network.
By controlling for these variables, the regression analysis aims to examine the relationship between cross-industry loan diversification and systemic risk, while considering the influence of various factors on the systemic risk level of banks.

Descriptive statistics
Table 1 displays the descriptive statistics of all variables in the main model for the period of 2007 to 2019, including loan diversification and systemic risk for 33 banks.The statistics consist of sample size, mean, standard deviation and quartiles.The average CoVaR is 2.362, with a median of 2.331, suggesting that the contribution of Chinese banking sector to systemic risk is slightly higher compared to European and American countries.The 6 Geographical diversification data are sourced from the former China Banking Regulatory Commission website, which provides detailed information on locations of bank branches.As of December 31, 2019, there were a total of 115,498 branch records of domestically headquartered listed bank holding companies (BHCs) available for querying.Using the STATA software, the city administrative units corresponding to each commercial bank branch are extracted based on the branch address information.In cases where branches lack specified city information, manual inquiries are conducted based on relevant sources to obtain the required data.
Based on the year of issuance of each branch's financial licence, an unbalanced panel dataset of bank branches at a semi-annual level is created, comprising a total of 1,576,791 records.The number of branches for each listed bank holding company in each city is then aggregated to construct the geographic diversification within the bank's branch network.
mean of 1-HHI is 0.842, with a median of 0.855, indicating that banks have a relatively high level of loan diversification, spanning across various industries.
In Table 2, the Pearson correlation coefficient matrix presents the relationships between all variables.CoVaR and 1-HHI exhibit a significantly positive correlation, indicating that cross-industry loan diversification exacerbates the banks' contribution to systemic risk.This suggests that as banks diversify their loan portfolios across industries, it leads to an increase in systemic risk in the banking sector.

Baseline empirical results
The baseline empirical results are presented in Table 3, where the impact of bank loan diversification (1-HHI) on bank systemic risk (CoVaR) is examined.Columns (1) to (2) present the results obtained using the Ordinary Least Squares (OLS) method.The coefficients of 1-HHI are found to be significantly positive, indicating that the cross-industry loan diversification increases bank systemic risk.To further control for time and individual effects, we utilise the bank-half-year fixed effect model, and columns (3) and ( 4) present the regression results.The coefficients of 1-HHI remain significantly positive, and after incorporating control variables, the coefficients become even higher and more significant, indicating a strong positive relationship between cross-industry loan diversification of banks and systemic risk.
In conclusion, the study's findings suggest that the diversification strategy aimed at reducing bank risk does not does not lead to a decrease in systemic risk.Instead, it exacerbates systemic risk within the banking sector.

Endogeneity tests
To address the potential endogeneity issues, we employ the Generalized Method of Moments (GMM) to further investigate the model.Due to the high autocorrelation of bank systemic risk, where the current level of systemic risk may be influenced by past values, and the possibility of endogenous relationships between systemic risk and bank-  specific variables, we introduce the first lagged dependent variable into the baseline model.This approach establishes a dynamic panel data econometric model, providing robustness testing to validate the findings.The specific approach is as follows: L.Systemic Risk represents the first lagged term of bank systemic risk.The coefficient β 1 associated with the lag term represents the speed of convergence to equilibrium, with values ranging from 0 to 1.In the context of dynamic panel data econometric models, two main estimation methods are commonly used: Difference GMM (DIF-GMM) and System GMM (SYS-GMM).To ensure the robustness and reliability of the research findings, both methods are employed in the regression analysis, providing a more comprehensive and accurate analysis of the relationship between cross-industry loan diversification and systemic risk in banks.
Table 4 presents the regression results of the dynamic panel data model in columns (1) to (4).Columns ( 1) and ( 2) represent the regression results using the Difference GMM method, while columns (3) and ( 4) use the System GMM method.Columns (1) and (3) utilise robust standard errors for estimation.The results show that the p-values of AR (1) test are all below 0.1, rejecting the null hypothesis, indicating that the residual terms have first-order autocorrelation.The p-values of AR (2) test are all above 0.1, accepting the null hypothesis indicating the absence of second-order autocorrelation in the residuals.The p-values of the Hansen test are all above 0.1, indicating the inability to reject the null hypothesis of instrument validity.This implies that the instrument selection is reasonable, validating the soundness of the model specification.The coefficient of the first-order lag term of systemic risk (CoVaR) is significantly positive, implying that the bank systemic risk has obvious inertia characteristics.Importantly, the regression coefficient of cross-industry diversification (1-HHI) remains significantly positive, which indicates that the relationship between cross-industry loan diversification and systemic risk is robust and not dependent on specific econometric modelling approaches.Therefore, the conclusions drawn from this study are robust and reliable.
To address the issue of self-selection, we employ the Propensity Score Matching (PSM) method to match the regression samples and re-estimate the baseline model.Specifically, firstly, the sample is dividing into two groups based on the degree of loan cross-industry diversification: highly diversified and lowly diversified.The highly diversified group is considered the treatment group, while the lowly diversified group serves as the control group.Secondly, we estimate propensity scores using logistic regression, and then employ two matching methods, namely k-nearest neighbour matching and radius matching, to match the propensity scores.The average treatment effect is calculated.The k-nearest neighbour matching method adopts the commonly used one-to-one matching method in micro company finance research, resulting in a closer approximation and smaller bias between matched samples.The radius matching method selects a radius of 0.01, which is commonly used in the literature.Finally, fixed effects tests are conducted on the propensity score matched sample, and the results are summarised in columns ( 5) and ( 6) of Table 4.The coefficient of cross-industry loan diversification (1-HHI) remains consistently positive and significant, consistent with the main regression results.This further confirms the reliability of our research findings.

Exclusivity explanations test 4.2.2.1. Geographical diversification.
Due to the close association between the geo- graphical diversification and the cross-industry diversification, in order to further eliminate the influence of geographic diversification to systemic risk, we conduct a more comprehensive test on geographical diversification.In the baseline model, the geographical diversification is constructed based on the number of bank branches at the prefecture-level cities, which only provides a partial depiction of banks' geographical diversification strategies from the perspective of branch quantity.To address this limitation, we constructed the geographic diversification indicator from the following perspectives, as suggested by Wang and Li (2021): 1)Weighted Distance (Distance).Using the data utilised in the calculation of 1-HHI G for the geographical expansion of bank branches, we make further adjustments to the HHI G index and calculate the weighted distance from the headquarters of listed banks.The specific calculation is as follows: Where, X i /X represents the proportion of branch institutions in the i city for the commercial bank, d i is the distance between the i city where the is the branch is located and the city where the headquarters is located, and N is the number of cities where the bank has branch institutions.This indicator reflects the geographical scope of listed commercial banks' geographical expansion.The greater the indicator, the greater the geographical scope of banks' geographical expansion, reflecting the breadth and extent of banks' geographical expansion strategy.
2) Geographical loan diversification (Geo_loan_diversity).To measure the geographical expansion of bank loans, we construct a loan diversification indicator based on the regional loan amounts reported in bank financial reports.Since the classification standards for loan sizes in the financial statements of listed banks vary by region from the CSMAR database, we aggregate the sample loan amounts based on cities or provinces and consolidate them into seven administrative regions nationwide.These administrative regions include East China, Central China, South China, North China, Northwest China, Southwest China, and Northeast China.Although the regional breakdown in the loan data is limited and only covers seven major regions in the country, our approach involved obtaining the original cross-regional bank loan data for each listed bank during the study period.The loan amounts in each of the seven regions for each bank in each half-year are used to calculate the HHI Loan index of the bank's regional loan distribution, resulting in geographical loan diversification indicator (Geo_loan_diversity).
These two newly constructed indicators are included as control variables in the regression model, replacing the original geographic diversification indicator.The empirical results are presented in columns (1) to (3) of Table 5.We find that the geographic diversification indicator is positively correlated with systemic risk.After including different dimensions of geographic diversification indicators as control variables, the coefficient of cross-industry loan diversification remains significantly positive in the regression model.These empirical results are robust, indicating that cross-industry loan diversification remains an important factor in exacerbating bank systemic risk, even after controlling for the effects of geographic diversification.

Individual operation cost. The implementation of bank loan cross-industry diversification may bring the following issues to individual banks:
Table 5. Regression results of exclusive interpretation.

Geographical diversification
Individual operation cost (1) Agency problems: The increase in industry clients enlarges the bank's loan portfolio, leading to a certain degree of expansion in the number of newly established institutions and the loan size of branch institutions.This can exacerbate the agency problems within the bank's organisational structure (Xia, 2016, Zhang, 2019).(2) Business complexity: Different industries have varying preferences for loan maturity structures and repayment methods.For example, the construction period for investment projects in the manufacturing industry is generally longer, and repayment starts after project completion.Additionally, different industries can provide different types of collateral for loans.Light-asset industries tend to prefer credit loans, while heavy-asset industries lean towards mortgage loans.These differences in providing heterogeneous services tailored to different business natures can make the bank's loan business more complex.
(3) Operational costs due to learning effects: The increased information asymmetry and the need for expertise in assessing and controlling loan risks across different industries can result in higher operational costs for the bank.
The above issues all contribute to additional operational costs for banks when diversifying into different industries.To control for the possible contribution of bank operating costs to systemic risk, we include the bank's cost of operation (Cost_of_operation) as a variable.
The operating cost indicator is measured by three variables: operating cost ratio (operating expenses/operating income * 100%), operating cash ratio (cash paid for other operating activities/operating cash outflows * 100%) and bank management expense ratio (bank management expenses/operating expenses * 100%).These indicators are included in baseline model for regression analysis respectively, and the empirical results are presented in columns (4) to (6) of Table 5.The results show that the coefficients of the bank's operating cost variables are not significant, while the coefficients of cross-industry loan diversification remains significantly positive.These results indicate that the increase in bank operating costs is not related to the exacerbation of bank systemic risk.Therefore, we can conclude that the mechanism through which bank operating costs contribute to the worsening of bank systemic risk have been excluded.

Redefine of systemic risk and control variables
In order to ensure the robustness of the research conclusions, we have redefined the measurement for bank systemic risk, as it is crucial for the findings of our study.In baseline model, ΔCoVaR measures the individual bank's contribution to systemic risk, which is the difference between the risk value of the banking system in an extreme crisis state and its value in a normal state.Following the variable settings proposed by Chu et al. (2020) and Guo and Zhao (2017), we set the extreme crisis quantile at the 5th percentile (q 1 = 0.05) and the normal state at the median (q 2 = 0.5), and calculate the VaR spread.When Adrian and Brunnermeier (2016) proposed the ΔCoVaR method, they suggest the value of q 1 represented the extreme crisis state, adjusting the value range of q 1 under the extreme state could effectively observe the contribution level of individuals to systemic risk under different tail risks.For example, Chen et al. (2015) set the extreme risk level q 1 as the 1st percentile and the 5th percentile to examine the risk value of financial institutions and interbank contagion effects under different extreme levels.Considering that China's financial prevention system and regulatory framework are still being improved and reformed, results indicate the robustness of the main hypothesis, supporting the conclusion of this study.

Sample period adjustment
In the baseline regression analysis of this study, the sample period covers the years 2007-2019.Considering the occurrence of the financial crisis in 2008 and 2009 during the sample period, significant changes occurred in the macroeconomic environment and financial system at that time.Particularly after the financial crisis, there has been increased attention from the government in preventing systemic risks and supporting the implementation of banking diversification strategies.Therefore, to account for these changes, the sample period in this study is adjusted to the post-crisis period (2010-2019) to examine the impact of bank loan diversification on systemic risk.In table 7, the resarch findings show that the main regression results remain substantively unchanged even after adjusting the sample period to the post-crisis period.

Mechanism tests: asset similarity channel
The above research confirms that as the cross-industry loan diversification increases, the bank systemic risk also rises.The specific implementation process of bank loan diversification may have the following issues: (1) Agency problems.
(3) Operating costs due to learning effect.These factors may result in additional operating costs for banks engaging in cross-industry diversification.However, we do not believe these issues would significantly impact systemic risk.Firstly, the increased costs incurred by banks in governance and operations could be offset by the positive economic benefits generated by cross-industry diversification.Secondly, the risks associated with governance, liquidity, and operations stemming from these issues can be mitigated and diversified through internal resource allocation within banks, thereby remaining as non-systemic risks to individual banks.
In line with our theoretical expectations, the channel through which might play a role likely lie in the increased likelihood of banks holding similar assets due to loan diversification.The increased similarity in bank asset holdings can lead to indirect contagion and significantly raise systemic risk through price losses (Chu et al., 2020;Duarte & Eisenbach, 2015, Fang & Zheng, 2016).Studies by Wagner (2010Wagner ( , 2011) ) and Allen et al. (2012) have shown that diversification strategies can result in similar asset structure, and banks with similar assets can be subject to common shocks, leading to joint liquidation and asset price discount.These affected banks may be forced to sell assets at a discount, generating negative externalities and vulnerabilities.Therefore, we specifically examine whether the assert similarity serves as a channel linking bank diversification to systemic risk (the asset similarity channel).
According to Girardi et al. (2021), a cosine similarity variable is constructed as a proxy for interbank asset similarity.Specifically, we construct the asset similarity (Cos AB ) between Bank A and Bank B as follows: Where, Ã is the n-dimensional vector (A 1 ,A 2 ,. . .A n ), A 1 ,A 2 ,. . .A n is the different asset items in bank A's balance sheet, each item is divided by the total assets and expressed as a percentage.B is B bank vector, defined the same way as bank A. Ã � j j B � � � � � � is the length of vector A and B. Cos AB is the cosine function between vector A and B. The value of this variable ranges from 0 to 1.The closer the cosine function is to 1, the more similar the two vectors are and the higher the similarity of the assets of the two banks.A cosine function of 0 means that the asset structure between two banks is completely different.We calculate the cosine similarity of all banks in the sample.The asset items include interbank deposit, trading financial assets, net buyback agreements, net loans and advances, and net intangible assets.The bank's asset similarity (Asset_simi) represents the average similarity between a single bank and all other banks.We also calculate the weighted average asset similarity (Asset_simi_w), which is the weighted average cosine similarity between bank i and other sample banks, calculated as follows: Where W j is the weight of bank j, defined as the proportion of j's total assets to the total assets of all banks.
To examine whether bank loan diversification leads to increased similarity in asset structures, we replace the dependent variable in Model (1) with the asset similarity index.Both the fixed effects model and OLS model are used to control for time effects and individual effects.The regression results are presented in Table 8.In columns (1) and (2), the coefficient of 1-HHI is 0.072 and 0.038, significant at the 10% level, indicating that cross-industry bank loan diversification leads to the increased asset similarity among banks.Banks with similar assets are more susceptible to common external shocks, which can result in joint liquidation and asset price depreciation, exacerbating bank systemic risk.Columns (3) and ( 4) show the results based on the fixed effect model.The coefficients of 1-HHI are 0.093 and 0.106 respectively, which are significantly positive at the 1% level.The consistent results between columns (1)-( 2) and ( 3)-( 4) further highlight the role of the asset similarity channel in the relationship between bank loan diversification and bank systemic risk, as an important mechanism contributing to the exacerbation of systemic risk.

Additional tests: impact of cross-industry loan diversification on individual non-systemic risk and overall risk
The above research finds that cross-industry loan diversification exacerbates the systemic risk, primarily through the increased asset similarity among banks.To provide a more comprehensive reflection of the economic consequences of bank loan diversification, we further examine its effects on individual banks' non-systemic risk and overall risk.Specifically, for reference Fan et al. (2010) and Wang and Li (2021),we adopt the nonperforming loan ratio of banks as the proxy for the non-systemic risk, and the sum of nonsystemic risk and the marginal contribution of systemic risk at the micro individual level for overall risk.
In Table 9, Panel A reports the results of the impact of cross-industry loan diversification of on individual non-systemic risk.Column (1) presents the result of the OLS model, while column (2) presents the results of the fixed-effect model.The empirical results from both models indicate that bank loan diversification does not have a significant impact on non-systemic risk, providing no evidence that asset diversification contributes to a reduction in individual banks' non-systemic risk.
Panel B reports the results of the impact of cross-industry loan diversification on the overall risk.In column (1), the coefficient of cross-industry loan diversification indicator is 0.958, significant at the 1% level, indicating that cross-industry loan diversification significantly exacerbates the overall risk.In column (2), the coefficient of bank loan crossindustry diversification is 1.871, significant at the 1% level, consistent with the result form OLS model.This further demonstrates the significant impact of bank loan diversification on overall risk and highlights the mechanism through which asset similarity leads to the deterioration of bank systemic risk.

Cross-sectional tests
The above research finds that, on the whole, cross-industry loan diversification increases systemic risk.We are interested in whether this effect varies under different circumstances, specifically considering the heterogeneous characteristics of the impact of credit diversification on systemic risk within the unique institutional background of China.
First, compared to European and American countries, bank lending in China is heavily influenced by national economic policy orientations.Policy influences can restrict crossindustry loans on one hand while promoting strategic orientations towards specific industries on the other.In China, apart from policy banks, banking institutions can be classified into state-owned banks, commercial banks, rural cooperative banks, urban credit cooperatives, and rural credit cooperatives.Specifically for listed banks, banking institutions can be categorised as joint-stock commercial banks and city commercial banks, which is also the common practice in existing literature (Li et al., 2020).Jointstock commercial banks are generally larger in scale with abundant loan resources, and among city commercial banks.Therefore, the overall impact of cross-industry loan diversification on systemic risk is smaller for city commercial banks.Second, in addition to policy influences, another important characteristic of Chinese banks compared to Western countries is the close strategic cooperation among banks.Bank strategic cooperation typically includes strategic partnerships between branches of joint-stock banks and city commercial banks in the same region and cooperation between commercial banks with similar strategic positioning in terms of resource sharing and joint market development.Banks involved in strategic cooperation engage in deeper collaboration in multiple business areas, strengthening the interconnectivity among banks.Compared to banks without strategic cooperation, these banks may be more inclined towards business diversification and have a more diverse approach to lending.In light of this, we expect that the risk effects of diversification may differ between banks with and without strategic cooperation.To better measure the extent of strategic cooperation among banks, we use interbank deposits as a proxy for measuring the level of strategic cooperation among banks.We then divide the research sample into two groups based on high and low levels of interbank deposits, and the regression results for each group are presented in columns (3) and (4) of Table 10.Through the fixed effects model, we find that when banks engage in cross-industry loan diversification, their contribution to systemic risk is more significant in the group with higher levels of interbank deposits, while the group with lower interbank deposits does not exhibit a significant effect.This indicates that banks with deeper cooperation and closer strategic relationships, resulting in higher similarity in business operations and asset structures with other banks, contribute to higher systemic risk through their credit diversification.
Finally, due to the high dependence of local governments on 'land finance' in China, there are strict restrictions and requirements on the proportion of real estate loans at the national level, aiming to avoid the concentration of largescale loans to real estate companies by individual banks.However, the demand for real estate loans continues to grow, leading to most banks inevitably being involved in lending to new clients in the real estate sector.The high leverage nature of the real estate industry may have important implications for the asset similarity and risk consequences in the banking industry.Therefore, we divide the sample into high and low real estate loan groups based on the average proportion of real estate loans.The regression results for each group are presented in columns ( 5) and ( 6) of Table 10.We find that the coefficient for 'diversification systemic risk' is higher and more significant in the high real estate loan group compared to the low real estate loan group.These empirical results indicate that a high proportion of high-risk real estate loans in the banking industry exacerbates the systemic risk externalities of diversification.Although government credit endorsement serves as a 'risk insulation' for the real estate industry, from the perspective of banks' risk management, in order to prevent systemic risk arising from asset similarity among banks, banks should cautiously expand their lending to high-risk areas such as real estate and adjust the proportion of high-risk real estate loans downward, mitigating the systemic risk caused by excessive credit diversification in the banking industry.

Conclusion
The issue of whether banks' adoption of diversified strategies increases systemic risk in the banking system is an important topic of both theoretical and practical significance.This study takes the perspective of cross-industry diversification in banks' traditional lending business and examines the relationship between cross-industry diversification and systemic risk in the banking sector, specifically focusing on the 'diversification systemic risk' issue in listed banks.The study finds a positive correlation between crossindustry loan diversification and bank systemic risk.The higher the degree of crossindustry loan diversification, the greater its marginal contribution to systemic risk.This 'diversification systemic risk' conclusion holds even after a series of robustness tests.Further research reveals that the impact of cross-industry loan diversification on systemic risk operates through the channel of asset similarity among banks.The higher the similarity in asset structures among banks, the greater the systemic risk.Moreover, the influence of cross-industry loan diversification on systemic risk varies among banks of different types, with different levels of interbank deposits and real estate loans.This effect is particularly prominent in joint-stock commercial banks, banks with higher levels of interbank deposits, and banks with higher levels of real estate loans.This study enriches the theories of diversification strategy and financial systemic risk, provides empirical evidence for bank risk management and business expansion, and offers theoretical and empirical support for optimising the scope of bank business, admission systems, and risk supervision by China's banking regulatory authorities.
However, this study has limitations.Due to the constraints of systemic risk indicators and data availability, the research focuses on listed banks as the sample.Undoubtedly, listed banks have systemic importance and can to some extent measure the overall risk level of the banking industry.However, it is also acknowledged that some non-listed banks may also have systemic importance, and the spillover effects of their risk may not be accurately measured in the design of the research variables.This calls for further research.
The findings of this study have important policy implications for risk prevention in the context of accelerating interest rate marketisation.In intense bank competition, the diversification strategy of banks may reduce individual risks but not the asset similarity and spill-over risks caused by diversification, leading to an increase in systemic risk.
First, banking regulatory authorities should effectively identify the systemic risk associated with diversified banking and further improve China's systemic risk prevention system to proactively address potential risks.When excessive systemic risk is identified in diversified banking, banks should be guided towards differentiated competition, leveraging their respective strengths and advantages to enhance their competitiveness in the market.
Second, in the context of interest rate marketisation, it is necessary to guide banks to compete in an orderly and rational manner, preventing excessive bank competition that could result in higher bank risk and jeopardise the stability of the banking system.China's banking sector is currently undergoing business transformation, with a high reliance on traditional businesses.Accelerating bank competition at this stage may lead to increased diversification in traditional banking, resulting in higher asset similarity and systemic risk.Therefore, during the ongoing financial system reforms, it is crucial to implement reforms related to bank competition gradually, opening up the competition in a controlled manner to prevent potential systemic risks in the banking industry.
Third, it is crucial to establish appropriate banking regulatory rules to prevent banks from engaging in regulatory arbitrage and evasive behaviours due to stringent regulations, as these actions could lead to an increase in systemic risk.This is particularly relevant for large joint-stock commercial banks and banks with high levels of strategic cooperation, as they have strong systemic importance and higher levels of risk spillover.Implementing moderate banking regulatory policies and rules would be more conducive to the operation and development of these banks, achieving the desired policy effects.
Finally, with the implementation of new policies by the central bank in the real estate sector, such as the dynamic adjustment of interest rates for first-home mortgages and the establishment of mechanisms to promote market transactions, the demand for funds in the real estate industry will continue to rise.Given the high leverage nature of the real estate industry, indirect financing risks have become prominent.Excessive funding support from banks through loan financing channels to the real estate industry will increase asset similarity among banks.To prevent systemic risk in the banking industry, it is necessary to establish multiple financing channels to alleviate the funding pressures of real estate enterprises, support leading companies in the industry to merge and acquire distressed projects, and effectively manage risks associated with troubled real estate enterprises.
Figure 1.Annual trend of cross-industry loan diversification of listed banks in China from 2007 to 2019. 4 at the 10% level.

Table 2 .
Correlation matrix of variables.

Table 3 .
Regression results of cross-industry loan diversification and bank systemic risk.

Table 7 .
Regression results of cross-industry loan diversification and bank systemic risk in post-crisis period.

Table 8 .
Regression results mechanism tests: asset similarity channel.

Table 9 .
Impact of cross-industry loan diversification on individual non-systemic risk and overall risk.