Banking on the Confucian Clan: Why China Developed Financial Markets So Late

Over the past millennium, China has relied on the Confucian clan to achieve interpersonal cooperation, focusing on kinship and neglecting the development of impersonal institutions needed for external finance. In this paper, we test the hypothesis that the Confucian clan and financial markets are competing substitutes. Using the large cross-regional variation in the adoption of modern banks, we find that regions with historically stronger Confucian clans established significantly fewer modern banks in the four decades following the founding of China’s first modern bank in 1897. Our evidence also shows that the clan continues to limit China’s financial development today.


2
The literature on financial development offers insights into the types of institutions that drive financial progress (La Porta et al., 1997, 1998Rajan and Zingales, 1998).
However, it remains unclear why some countries had the necessary institutions in place and ready for financial development to take off, whereas others did not. 1 To answer this question, we need to go back further in history. Using both theory and anecdotal evidence, Greif and Tabellini (2017) claim in a recent major contribution, that during the last millennium the West has relied on the 'corporate entity' or corporation to achieve interpersonal cooperation, whereas China has relied on the Confucian clan.
The West has long emphasised cooperation beyond bloodlines, seeking to develop nonkinship organisations (e.g. the Church, corporations and non-profit organisations) and supporting impersonal institutions. As a result, when the demand for resource pooling and risk sharing among strangers grew exponentially after long-distance ocean trade began to take off in the 16 th century, the Dutch and British had the institutional capacity to develop large-scale public capital markets and arms-length finance. In contrast, from the Han dynasty (206 BCE-220 CE) onward, 2 China focused on perfecting lineage institutions and kinship morality to achieve cooperation among relatives. Therefore, the divergent institutional-development trajectories in China and the West date back to choices made 1,000 years ago regarding the mode of interpersonal cooperation.
In this paper, we explore the implications for financial development of the divergence in institutional-development trajectories between China and the West, using historical and contemporary data. We show that the deeply entrenched Confucian clan has historically stifled and continues to depress both the demand for and the supply of external finance in China. It is helpful to think of the Confucian clan as an internal financial market for its members in which resource pooling and risk sharing are neither based on explicit financial instruments or agreements nor enforced by the rule of law. Rather, they are based on implicit contracts that dictate the rights and obligations of each member with respect to other members and are enforced by Confucian rules. These characteristics have had two important consequences. First, Chinese people had little demand for external finance before the Industrial Revolution, because the Confucian clan and financial markets were functionally substitutable.
Second, the focus on the biologically based clan for resource pooling led the Chinese 1 La Porta et al. (1997) demonstrate that the historical development of a country's legal institutions affects its financial market development. Haber (1991) shows that political and national institutions play a critical role in financial development. The cause of the financial revolution in England in around 1700 is still widely debated (Temin and Voth, 2008). 2 For reviews of the history of Confucianism and its spread in China, see Ebrey (2014) and Lang (1946). 3 elite to neglect the development of the impersonal institutions needed for formal armslength finance. Lacking these institutions, China could not supply external finance even when the need arose as a result of the late 19 th century industrialisation efforts. This explains the country's sluggish adoption of modern finance in the late 19 th century (Kirby, 1995;Goetzmann et al., 2007). China's early adoption of Confucianism as the state orthodoxy to structure society for interpersonal risk sharing and resource pooling made the country both uninterested in and unprepared for later external financial development.
For our empirical tests, we use cross-regional banking data for the  period. Modern banking was launched in China in 1897 as part of the country's modernisation movement, when the institution of the Confucian clan had reached its zenith. 3 During this period, the 283 prefectures in our sample varied both in their adoption of modern banking and in the prominence of the clan, offering an ideal context in which to test our hypothesis about the clan and financial markets being substitutes.
We measure each prefecture's clan strength using genealogy density, defined as the number of genealogy books per 10,000 population that had been compiled as of 1897.
Genealogy books were written records of the lineage's male members since its inception. Importantly, they also contained clan rules transmitting Confucian teachings and prescribing a code of conduct for members, usually detailing rewards for conformity and punishment for violation. The ultimate purpose of these books was to promote clan solidarity and secure intra-clan intertemporal exchange. Our analysis finds that in regions with more dense genealogical records (genealogy), i.e., stronger Confucian clans, significantly fewer modern banks were established during the 1897-1936 period.
For 1936, we find that doubling genealogy density reduces bank density by 5.6%. This suggests that strong clans impeded financial development.
Our key explanatory variable, clan strength, may proxy for omitted variables correlated with the development of modern banks. For example, historically impoverished regions may have developed greater reliance on the Confucian clan, and it might not have been profitable for modern banks to expand into poor regions. To address this concern, we instrument clan strength using a prefecture's shortest greatcircle distance to the nearest academy at which the Confucian master Zhu Xi (1130-1200 CE) taught. Zhu Xi played a crucial role in spreading the Confucian way of life at the grassroots level by encouraging every clan to build an ancestral hall. This was the centre of clan life, where all male members gathered at least once every three 3 The first modern Chinese bank, the Imperial Bank of China (renamed the Commercial Bank of China in 1912), was founded in 1897. The second, now known as the Bank of China, was founded in 1904 (Wu, 1935). 4 months to conduct ancestor-worship rituals. Between 1167-1194 CE, Zhu Xi frequently lectured on his operationalised version of Confucianism at three academies. Most of his 448 disciples were from prefectures that were close to these academies (Chan, 1982).
Accordingly, his influence on nearby regions is expected to have been high.
Furthermore, as these academies were located in neither economic centres nor remote towns, a region's distance from the nearest academy had nothing to do with regional economic conditions or accessibility. Consequently, the academies' impact on banking development 600 years later should only be through Zhu Xi's indoctrination and its continuing effect on local clan efforts. This exclusion restriction is confirmed by our randomisation inference tests, in which we construct counterfactual spatial configurations of the academies at which Zhu Xi taught and find that the distance to the counterfactual academies has little effect on the distributions of clans and modern banks. Our instrumental variable (IV) results reinforce our baseline conclusion; the negative effect of clan strength on bank development becomes approximately 1.25 times stronger.
To examine whether the clan effect is driven by local factors that are not controlled for in our covariates or IV analyses, we conduct a placebo test that shuts down the clan competition channel. We repeat our main analysis by studying the spread of foreign and official banks from 1903 to 1936. The expansion of foreign banks in China in this period was primarily driven by the need to serve foreign-owned businesses, and official banks were established to facilitate government activities. As the services provided by these banks did not compete with clan-based informal finance, they probably had little to do with local clans. Indeed we do not find a relationship between clan strength and the presence of foreign or official banks.
As the Confucian clan provided resource pooling services for its members, it is expected to have reduced their demand for, and hence depress the price of, external capital. Clan strength may also have reduced the cost of capital by strengthening local social order and trust, making intertemporal contracts more secure. We thus expect interest rates in regions with stronger clans to have been lower during the sample period. Indeed, based on private-loan data from the China Historical Interest Rate dataset in Chen et al. (2016), we find that interest rates during 1912-1936 were significantly lower in prefectures where clan structures were stronger, supporting our claim.
As Confucianism was at the heart of Chinese society for more than 2,000 years, it is appropriate to ask whether its teachings and value system continue to impact China's financial development today. Our final empirical analysis addresses this question of long-term persistence. Using data from the 2010 China General Social Survey (CGSS), we find that in prefectures with strong Confucian clans, people (1) 5 tend to trust relatives more and outsiders less, (2) prefer not to use formal contracts in business with relatives, and (3) are relatively unlikely to borrow money from banks.
Further, these prefectures have less per capita banking and a smaller banking sector relative to GDP. Taken together, these results suggest that Confucianism casts a long shadow on financial market development in China, with competition between the clan and modern banking ongoing today. Alesina and Giuliano's (2015) findings regarding how cultural institutions compete with and substitute for formal finance, we show that when clan solidarity is high, the influence of the clan-based order may dominate that of external financial markets. Our findings in China may also help to explain why formal financial development in Italy, Spain and Portugal has lagged behind that in the UK and the US. We suggest that after the Counter-Reformation in southern Europe (Becker et al., 2016), Italy and Spain retreated from financial development and sought to rebuild and strengthen their kinship clans for resource pooling and risk sharing, leading to the rise of the extended family and the stagnation or decline of formal finance. 4 However, the cases of Italy and China have an important difference: the retreat from finance starting in the 16 th century led Italy to reconstruct its clan institutions, whereas the excessive Confucian focus on the clan prevented the Chinese from developing an appetite for external finance. The advantages and disadvantages of intra-family risk sharing are studied in other contexts by Kotlikoff and Spivak (1981), Townsend (1994) and Alesina and Giuliano (2010), among many others. Formal financial institutions must contend with informal insurance and financing networks supported by local cultures.

Consistent with
Broadly, our study also contributes to the body of work on the effects of cultural and religious institutions on economic growth (e.g., Knack and Keefer, 1997;Guiso et al., 2003Guiso et al., , 2016Becker and Woessmann, 2009;Grosjean, 2011;Voigtländer and Voth, 2012;Giuliano and Nunn, 2017). 5 In a related study, de la Croix, Doepke and Mokyr (2018) show that the guild-based apprenticeship institution in pre-industrial Europe promoted knowledge diffusion beyond bloodlines and led to development. Following this line of reasoning, we can infer from our findings that China's closed Confucian-6 clan has constrained knowledge transmission beyond the kinship network and thereby hampered economic growth in historical China.
As said earlier, existing studies usually assume that societies had different institutions or cultures at the beginning of the modern era, with distinct development outcomes. Following Greif and Tabellini (2017), however, our study takes us back to when China and the West diverged in their institutional-development trajectories. At the dawn of the Industrial Revolution, China and the West had very different sets of institutions. Our empirical tests substantiate the claim that China's Confucian clancentred institutions are not conducive to external finance.
Our work also relates to studies of the role of Confucian culture in China's economic development (Kung and Ma, 2014;Chen et al., 2019). By documenting the negative effect of Confucianism on modern banking, our analysis contributes to the literature on the determinants of China's modernisation process in the late Qing dynasty (e.g. Goetzmann et al., 2007;Yuchtman, 2017). This paper has six parts. Section 1 discusses the economics of the Confucian clan.
Section 2 presents the data used in our empirical tests. Section 3 studies the impact of the clan on the spread of modern banking in the early 20 th century. Section 4 investigates whether Confucianism reduces demand for external finance. Section 5 discusses the long-term impact of Confucianism on China's financial development, and Section 6 concludes the paper.

Economics of the Confucian Clan
Like other traditional societies, China has relied on the extended family for risk sharing and resource pooling for many millennia. However, what makes the Confucian clan unique is not its functional structuring of intra-clan intertemporal exchange but its diverse institutional elements, which have evolved since the time of Confucius (551-479 BCE). These innovations include the codification of rules and norms in genealogy books; the embodiment of the clan through physical facilities (e.g., ancestral halls and ancestor tablets for worship); the introduction of regularised ancestor-worship rituals; the development of a lineage estate trust holding land and other property; the provision of relief for impoverished and otherwise less fortunate members; and the structuring of the clan as a business. 6 Combined with the original prescriptions formed by Confucius and his followers, these innovations transformed Confucianism from a loose collection 7 of norms into an institution that organised daily life, similar to monotheistic religions.
As a result, everyone's position in the stratified social hierarchy and his obligations to others were all as dictated by Confucian rules and remained unchanged for life. This constancy is what made such morality-based interpersonal resource pooling and sharing more secure than free will-based market exchange. 7 By the beginning of the period under examination , the Confucian clan had reached its zenith. It was a clearly regulated property ownership system in which the husband owned property rights in and of his wife and, likewise, the father owned property rights in and of his children; furthermore, The head of the family was its eldest male member. … He held the title to all family property and he alone could dispose of it, as well as of the earnings and savings of all the family members. He settled the marriages of his children and signed the marriage contracts. … Furthermore, the law exonerated the father or grandfather who killed his son or grandson unintentionally when chastising him 'in a lawful and customary manner'. Nobody disputed the right of the head of the family to sell its members into slavery. (Lang, 1946, 26-27) This system of property rights was supported by filial piety, the highest moral principle in Confucianism. This ensures that children, in the eyes of parents, are secure and personalised instruments of investment and insurance for old age and sickness. To place our empirical work in context, we first highlight four aspects that made the Confucian clan a robust internal financial market for resource pooling and sharing.
This internal financial market was resilient enough to fend off initial competition posed by formal finance during China's modernisation.
First, the rules guiding the bilateral rights and responsibilities of any two members of a Confucian clan were determined by their biological distance, generational gap, relative age and gender. In general, a smaller biological distance between members translated into a greater obligation to share property and income. A kinsman belonging to an older generation or who was older had more claim on the junior's property and income, whereas the latter bore more responsibility to care for the former. Likewise, a man had more seniority than a woman in entitlement claims (Lang, 1946;Ebrey, 2014).
Second, from the Song dynasty onward, clans became prominent owners of land and other property, the income from which was mostly used to pay for lineage festivals, weddings and funerals, children's education, rewards for success, disaster relief and assistance for poor clansmen, and to provide loans for business and emergency needs 7 For more on this, see the references cited in the preceding footnote. 8 of clan members. A prominent Confucian scholar-official, Fan Chung-yen (989-1052 AD), created the Fan Clan Charitable Clan Estate in 1050, donating 1,000 acres of his own land as a permanent reserve. He also wrote a charter detailing the estate's operational rules and emphasised the need to help the poor and unfortunate among the Fan clan members (Twitchett, 1959). This model of the charitable estate was widely replicated across the country until the 20 th century (Feng and Yan, 2012). 8 The resulting charitable estates not only gave the Confucian clan a tangible presence and elevated its risk sharing function, but also boosted its appeal to members and increased clan solidarity.
Third, under Confucianism, the formal code recognised property rights as held at the level of either the nuclear family or the tang (Zelin, 2005), rather than at the individual level. A tang (literally the short version of the Chinese term for 'ancestral hall') was a fictional corporate entity. For business purposes, the clan was treated as the equivalent of a Western corporation, owning land and other assets and conducting business on behalf of clan members. 9 For example, a group of clan members could pool assets and set up a tang in the name of a deceased common ancestor. A clan could have just one tang with all clan members as its shareholders, or it could have multiple tangs located at different nodes along the clan tree. Clan members were expected to support other members in starting businesses or during times of need. Secured by Confucian norms and clan rules, these tangs were instrumental for intra-clan fund raising for business ventures, providing a trusted form of internal financial intermediation for economic growth in pre-modern China (Faure, 1989;Shiroyama, 2004;Zelin, 2005). 10 However, as unlimited-liability organisations, tangs were largely unable to pool resources and spread risk beyond the clan's own membership, limiting their ability to scale up and compete with modern corporations in the industrial age (Kirby, 1995). 8 For example, during the Qing dynasty, a Confucian scholar-official named Zhang Zhao established a 'coffin home' fund (zhangshi yizhuang) for his clan in Songjiang prefecture (where Shanghai is located today). He donated 1,000 mu (approximately 164.8 acres) of land to the coffin home fund. The land was put to productive use, and the revenue was used to support clan members in need. 9 After the British came to Malaysia and Hong Kong in the 19 th century, a number of lawsuits occurred concerning whether a Chinese tang should have the same standing as a 'legal person' in a British court as a corporate entity. The British court ruled that it did not, forcing most Hong Kong tangs to re-incorporate in the 1920s and 1930s. See Chung (2010 Finally, prior to the Song dynasty, genealogy and ancestral halls were the exclusive privilege of the aristocratic class. However, starting in the late Song dynasty, both practices were popularised to the grassroots level, with the genealogy expanded to include clan rules operationalising Confucian principles, prescribing a code of conduct and stipulating corresponding rewards and punishments (Liu, 1959). With every clan permitted to construct its own ancestral hall, clan members had a physical place to gather and worship ancestors together at least once each quarter and during festivals. The ancestral hall was also where banquets for births, weddings and funerals took place. One key function of such regular gatherings was to allow members to gather information about each other and monitor kinsmen's behaviour. If they discovered free-riding, the clan's elders would issue early warnings, addressing the moral hazard that might arise from informal clan-based resource sharing. Thus, the genealogy and ancestral hall became two key devices for strengthening solidarity and building cohesion among clan members. Yet while some regions were quick to adopt both devices, others were less responsive. In this study, we rely on genealogy book data to measure the intensity of Confucian clans' influence in each region, given the lack of usable ancestral hall data across regions.
After more than two millennia of collective effort and development, the Confucian system centred around the clan worked reasonably well in pre-industrial China. It provided adequate resource pooling for business and risk sharing for everyday life and supported the largest economy in the world for hundreds of years (Faure, 1989(Faure, , 2006Zelin, 2004;Pomeranz, 2009). It was largely the only effective interpersonal cooperation medium that people could securely depend on. However, this success came at a high price: the Chinese intellectual elites had no incentive to develop the impersonal institutions needed for formal financial development (Kirby, 1995). For example, the Qing Dynasty Code had only 31 sections (out of a total of 2,354) devoted to commercial matters, treating them almost as a footnote (Peng and Lin, 2020). The Qing state largely left commercial and financial matters to the clans and grassroots associations (Cohen et al., 1981), which made the clan effectively the de facto enforcer of property rights and civil rules (unless a dispute led to someone being killed). 11 Furthermore, the success of the Confucian clan in meeting members' multiple needs also served to entrench it as an institution and made Confucian scholar-officials more interested in defending it against potential competitors, such as external financial markets. In sum, the long-term success of the Confucian clan as an internal financial market worked against the development of formal finance in two ways: it depressed the demand for and limited the supply of external finance.
After the Opium War of 1839-1841 revealed the power of Western technology and large-scale capital mobilisation, Chinese elites were forced to adopt Western financial institutions, such as modern banking, joint-stock limited-liability corporations and formal securities markets (Kirby, 1995;Goetzmann et al., 2007). This pitted the clan, as the incumbent facilitator of resource pooling, against formal arm's-length finance as the newcomer. The sections to follow use both the cross-regional results from that competition and recent survey data to test this claim.

Data Description
From the late 19 th century to the early 20 th century, while Confucian influence was still in its heyday, China saw the creation of modern financial markets. Modern banks were introduced by the British and other Europeans after the country was forced to open in 1842. In 1897, Chinese elites began to set up their own modern banks. This provides us with a unique historical context to examine the effect of Confucian culture on financial development. Another unique advantage of the 1897-1936 period is that financial development was primarily manifested through the rise of modern banking, rendering the measurement of financial development cleaner and more uniform than that of other markets at the time.

Banks
To measure financial development, we use the number of modern Chinese banks (and branches) operating in each prefecture in 1936 as our primary metric. After the first modern Chinese bank, the Imperial Bank of China (Zhongguo tongshang yinhang), was established in 1897, the banking industry followed a sustained growth trajectory and reached nearly 2,000 banks (including branches) by 1936, covering 73% prefectures across China proper (Figure 1). 12 We use 1936 as the end year because the Sino-Japanese war broke out in 1937, complicating the socioeconomic environment. The bank data are obtained from county gazetteers (xianzhi) and the Banking Yearbooks compiled between 1912 and 1937.
Specifically, following Stauffer (1922), we define bank density, banks, for each prefecture as the number of banks per 10,000 members of the prefectural population 11 in 1920, the last year prior to 1936 for which prefectural population data are available.
To attenuate the effect of outliers, our main measure is the natural logarithm of bank density (plus 1), i.e. ln(banks+1). To gauge the growth of banks between 1897 and 1936, we use the annual average number of modern Chinese banks operating in each prefecture between 1897 and 1936 as an alternative outcome variable. We normalise it by the average prefectural population between 1910 and 1920 and take the natural logarithm (plus 1).

The Confucian Clan
As stated by Tabellini (2010, 2017), the Confucian clan was the predominant medium of cooperation from the Song dynasty onward. To measure the strength of the clan in a prefecture, we use the number of genealogy books compiled in the prefecture before our period of analysis. Genealogy books not only recorded lineage members and relationships to solidify a sense of belonging and promote lineage cohesion (Bol, 2008) but also specified clan rules for members to follow, with 'carrots' detailed for conformity and 'sticks' for deviation (Liu, 1959). We can thus infer that regions that cared more about clan solidarity, and hence devoted more effort to organising clans, kept more genealogy books.
The genealogy data are obtained from Shanghai Library's (2009) Comprehensive Catalogue on Chinese Genealogy (Zhongguo Jiapu Zongmu). The Catalogue contains the genealogies of up to 52,306 clans, including more than 700 surnames across 283 prefectures, in China. After excluding incomplete records (missing location and/or compilation year) and records outside China proper, there are 41,370 clan records for use in our study. The majority of these genealogy books were compiled during the Ming and Qing dynasties . We use books compiled before 1897, the year of establishment of the first modern bank, as our primary measure of clan strength (clans). By doing so, we avoid any reverse causality from financial development on Confucian clans' strength. Similar to our construction of bank density, we calculate genealogy density as the number of genealogy books 13 per 10,000 members of the population in the prefecture (based on the population averaged over the 1393-1851 period). 14 The distribution of genealogy density is highly skewed, as a few prefectures have many more genealogical records than the others (Online Appendix Figure A1).
To mitigate bias from extreme values, we remove the top 5% tail (13 prefectures) of genealogy density and take the natural logarithm of this variable (plus 1). 15 The genealogy sample in the Catalogue may exhibit survivorship bias, as some genealogy books may have disappeared or been destroyed before 2009, and the survivorship patterns may differ from one prefecture to another. However, this survivorship bias should strengthen our proxy for Confucian influence because genealogical records were more likely to survive in prefectures that historically had strong Confucian adherence and powerful clans. Moreover, the geographic distribution of genealogy books (Figure 1) is consistent with historical narratives of the distribution of clans: southern and southeastern China (especially Zhejiang, Fujian, Anhui and Guangdong provinces) cared more about clans than northern and western China did (Zheng, 2001;Szonyi, 2002). This suggests that our genealogy records at least do not suffer obvious systematic survivorship bias in terms of geographic distribution. We return to this issue when discussing the IV approach in Section 3.

Geographic Controls
We control for the following geographic factors that may be correlated with both clan strength and financial development.
China's modernisation began with the treaty ports -the territorial concessions given to Western powers, beginning with the Treaty of Nanjing in 1842. Under Western influence, modern banks were established in the treaty ports before expanding to inland areas. Meanwhile, the Confucian ethos was gradually undermined by Western culture in or near the treaty ports. We control for the spill-over effect of the treaty ports using the shortest distance from a prefecture to its nearest treaty port established before 1897. Furthermore, given the institutional differences between the treaty port prefectures and the traditional inland prefectures, we remove from our analysis the 26 prefectures where treaty ports were established before 1897. Doing so also helps attenuate the outlier concern in bank data. For example, Shanghai, as a treaty port, 14 Cao (200014 Cao ( , 2017 provides estimates of prefectural population for the years of 1393, 1580, 1630, 1680, 1776, 1851, 1880 and 1910. We drop 1880 because the unprecedented population losses during the Taiping rebellion (1850-1864) had not been recovered by then. We also drop 1910 to avoid the feedback effect from bank development. 15 The results remain consistent when we use the full sample (284 prefectures) and hence are not reported separately. 13 had 187 banks by 1936, twice as many as the second highest prefecture and 33 times the sample mean ( Figure 1).
Historically, China had a considerable east-west divergence, with eastern China being more developed and prosperous. Accordingly, clans and banks were more concentrated and more developed in eastern China ( Figure 1). Thus, the effect of the Confucian clan on banking is subject to this inherent east-west difference. To address this concern, we include a control dummy for eastern China that takes a value of 0 for prefectures in the five western provinces (Gansu, Guangxi, Sichuan, Guizhou and Yunnan) and 1 otherwise. 16 In addition, we control for distance to the coast, distance to the nearest navigable river, and whether a prefecture was a provincial or national capital. Summary statistics for all the variables are reported in Online Appendix Table A1.

Confucianism and Modern Banking
Our empirical analysis focuses on the hypothesis that the Confucian clan competed with and impeded the development of modern banking in the late 19 th and early 20 th centuries, and that more modern banks were adopted in regions with a relatively weak clan presence. We first illustrate this pattern in Figure  To formally test the effect of the Confucian clan on banking, we estimate the following cross-sectional regression: where Banks i is the natural logarithm of (1 plus) the modern bank density in prefecture i in 1936, Clans i is the natural logarithm of (1 plus) the genealogy density compiled in prefecture i before 1897, and X i is a vector of geographic control variables as described 14 in sub-section 2.3. We use heteroskedasticity-robust standard errors. In addition, to deal with potential spatial correlation among the residuals, we cluster the standard errors within a radius of 136 kilometres based on Colella et al. (2019). We choose 136 kilometres because it is the average distance between two adjacent prefectural seats.
Clustering by this radius thus approximately covers the banks of adjacent prefectures. 17 The results are reported in Table 1. We first examine the univariate clan effect on banking (column 1), before controlling for other factors. The ordinary least squares (OLS) estimation shows that the Confucian clan has a significantly negative effect on bank density, and this effect becomes greater and more significant after controlling for geographic factors (column 2). The clan coefficient of -0.05 implies that doubling the average clan density for a prefecture decreases bank density by 5%.

Additional Controls
The distribution of clans and modern banks may be related to a gamut of other factors.
An obvious one is the need for finance arising from economic development. To the extent that clan formation was an endogenous response to needs, people in a region could respond to increasing demand for financial intermediation by either forming more clans and making them stronger or by developing more financial markets if this was feasible and the appropriate institutional infrastructure existed. For regions lacking economic growth opportunities, the need for more clans or more external finance would not arise. 18 This may partly explain why prefectures in the historically prosperous southern and southeastern China exhibit higher clan density than northern and western China (see Figure 1). We control for the impact of regional variation when assessing the need for resource pooling. 17 We also use greater radii, such as 204 kilometres that approximately cover the next ring of adjacent prefectural seats. This has little influence on the clan effect (not reported). 18 Greif and Tabellini (2017) point out that time variation in the national aggregate of genealogy books reflects political changes. The decision to form a new clan in a village was clearly endogenous and a function of need, feasibility and manageability (there was a limit to how large a clan could be before it became unmanageable). Therefore, the time-series variation for the national number of genealogy books should at least partly reflect the country's time variation in economic activity and risk event frequency (i.e., needs). An in-depth investigation of the exact functional relation between needs and clan formation is beyond the scope of the present paper, but we control for this effect through the inclusion of needs-related proxies.

15
During our sample period, financial needs were mainly driven by rising industrialisation and growing trade and commerce after China began to open up in 1842 (Rawski, 1989). We proxy for industrial development using each prefecture's number of modern industrial firms per 10,000 people in 1896. We also proxy for commercialisation using the total number of members in each prefecture's local chambers of commerce (per 10,000 people) in 1913. In addition, we control for each prefecture's urbanisation rate in 1910 to capture overall economic prosperity, as financing needs are higher in more prosperous areas.
It is also possible that the inaccessibility of remote regions caused both a decrease in banking, due to high costs, and an increase in clan activity, because residents were more likely to be bound together and forced to form strong clans. However, historical evidence suggests that regions dominated by clans (especially southern and southeastern China) were usually not closed or inaccessible but rather prosperous due to trade and market integration built on their comprehensive river and courier networks. Beginning in the late 19 th century, many regions were also connected by the telegraph, a communication medium that was instrumental in expanding banking at the time . To rule out this channel, in which clan strength simply proxies for inaccessibility, we not only include the geographic controls described earlier but also adjust for both the number of telegraph stations and the number of postal offices in each prefecture in 1896.
Protestant Christianity, which was arguably instrumental to capitalist growth (Weber, 1905), competed with Confucianism during our period of analysis. Protestant missionaries brought to China not only a new religion but also Western institutions and 'useful knowledge' for economic development (Bai and Kung, 2015;Chen et al. 2014 Regional clan strength is known to be associated with human capital achievements. To help sons compete in the imperial civil examination, 20 a resourceful clan would typically provide support by hiring tutors, acquiring books and paying for schools. Thus, regions with a strong clan culture enjoyed more jinshi (the highest imperial civil-exam grade) degree winners (Elman, 2013;Twitchett, 1959 As shown in Table 1, we gradually include the above-mentioned controls into our regressions in order to correct for other effects and rule out other channels. Specifically, column 3 controls for financial needs: density of industrial firms, commercialisation and urbanisation. Column 4 further controls for accessibility: the number of telegraph stations and post offices. Column 5 considers Protestantism, jinshi holders and warlord conflicts. The results confirm the baseline claim that the clan still had a significantly negative impact on bank density. Moreover, conditional on the full set of controls, the clan effect (0.056 in column 5) is now slightly greater than that with only the geographic controls (column 2), suggesting that the clan effect may be underestimated if we do not adjust for the many correlates and possible channels.
Given that 36% of the prefectures in our sample had no banks in 1936, we also run a Tobit estimation, with the results remaining robust (column 6 of Table 1).
Finally, we find a similar negative effect of the Confucian clan on banking when using the average number of modern banks each year between 1897 and 1936 (column 7). 21 The jinshi data are obtained from Zhu and Xie (1980). 22 As noted by Rowe (1998), during Chen Hongmou's tenure as the governor of Jiangxi province in the middle 18 th century, he greatly expanded the rule-making and enforcement power of the head of each clan, which helped to strengthen the clan's cohesion and solidarity. But, to the emperor's dismay, the strong clans also led to increased frequency of inter-clan conflicts, including the establishment of clan militias. There appeared to be a link between clan strength and interclan violence.

Instrumented Results
After controlling for various effects and ruling out several channels, as shown in Table   1, our estimated clan impact on banking may still be biased due to unobserved prefectural factors. For example, given that Confucianism was the state orthodoxy, prefectures with stronger Confucian clans may also have had more powerful local governments, inhibiting bank development. In addition, clan strength may be associated with strong general social capital, which is conducive to economic development (Guiso et al., 2004(Guiso et al., , 2008. Another factor to consider is measurement errors in the number of genealogy books due to survivorship bias, as discussed in Section 3 (i.e. the bias could underestimate the real effect of the clan on financial development). To address these possible concerns and further establish the Confucian clan as the main channel stifling banking, we additionally use each prefecture's shortest distance to the nearest Zhu Xi academy in the 12 th century as an IV. It is plausible that this instrument only affected modern banking growth 600 years later through its effect on regional clan strength, as explained below.
As mentioned in the introduction, Zhu Xi was the most important figure responsible for popularising Confucianism, especially the Confucian clan, in grassroot communities from the 12 th century onward. His philosophy developed and spread during his teaching at Yuelu Academy in Changsha (in Hunan province), Hanquan Academy in Jianyang (Fujian) and Bailudong Academy in Jiujiang (Jiangxi), as displayed in Figure 1. We refer to these as 'Zhu Xi academies'. 23 At these academies, he completed several classical texts, such as Rituals of the Family. These classics provided operational details on how the Confucian clan should be organised, and this organisational structure was widely adopted from the 13 th century to the 19 th century (Ebrey, 1991).
Meanwhile, Zhu Xi recruited students and invited renowned Confucian scholars to lecture at his academies (Shu, 2003). Between 1150 and 1200, 448 disciples studied under Zhu Xi. These disciples came from 58 nearby prefectures (Chan, 1982) (Online Appendix Figure A2), and they not only spread Zhu Xi's teachings across several regions but also reprinted his books and wrote commentaries from various perspectives (Chan, 1987). The curricula and learning methods of the Zhu Xi academies became the standard for both official and private schools in the succeeding dynasties. More importantly, Zhu Xi's books were required reading materials for the imperial civil examinations from the 14 th century to 1905. These various channels helped to diffuse his version of Confucianism and thus the Confucian clan (Shu, 2003).
Given   (Zhu, 1996). In 1193, Zhu Xi was 19 appointed a governor of relief in Changsha, Hunan province, where he seized the opportunity to introduce his philosophy and curricula to Yuelu Academy in Changsha (Shu, 2003). Given that his official appointments and mourning rituals were not chosen by Zhu Xi, the three academies were arguably 'selected' by exogenous forces.
Second, the Zhu Xi academies were located in neither economic centres nor backward regions. Accordingly, a prefecture's shortest distance to these academies does not explain its economic prosperity as measured by commercial tax and population density between 976 and 1102 (columns 1-3 of Table 2). Neither does distance predict long-term development as measured by each region's urbanisation, commercialisation and industrialisation in the early 20 th century (columns 6-8). In addition, the prefectures close to the Zhu Xi academies were neither remote nor less accessible, as evidenced by the low correlation between their distance to the academies and development of communication in the 19 th century (telegraph and post offices) (columns 9 and 10). However, distance to the academies has a significantly negative correlation with the density of jinshi degree holders during the Song dynasty (column 4). This suggests that the Zhu Xi academies might have had a spill-over effect on human capital in nearby regions at the time. However, this is not a serious concern, as this spill-over effect disappeared in the subsequent Ming-Qing period  (column 5), and we control for each prefecture's jinshi density in the estimations.
[ From this remaining subset, we randomly draw three counterfactual Zhu Xi academies and repeat this process 2,000 times, resulting in 2,000 sets of counterfactual academies. 27 Then, for each set, we use a prefecture's distance to the nearest counterfactual Zhu Xi academy to predict its clan density and bank density, respectively. This process creates 2,000 regression coefficients for each of the two outcome variables. Figure  The results show that distance to the counterfactual Zhu Xi academies has much less power to predict the spatial distribution of clans than distance to the actual academies. Evaluated by the p-value, the share of counterfactual coefficients that are greater than the coefficient based on the three true Zhu Xi academies is only 5.4% (from the sample of prefectures with an academy) or 3.6% (from the above-median jinshi sample). Instead, the majority of the counterfactual coefficients are close to 0 (Figure 4(a) and (b)), indicating that the geography of clans after the Song dynasty was indeed shaped by the cultural spillover of the Zhu Xi academies. Likewise, we find that the counterfactual distance has weak power in predicting cross-regional banking development in the early 20 th century, with only 4.5% to 6.7% of counterfactual coefficients greater than the coefficient based on the actual Zhu Xi academies (Figure 4(c)-(f)). Therefore, our IV is unlikely to affect financial development through channels other than the Confucian clans.
[ Figure 4 about here] In Table 3, after instrumenting by distance to the Zhu Xi academies, we still find that clan strength has a significantly negative effect on bank development (column 1).
The result remains robust when we additionally control for financial need, The F-statistics suggest that distance to the academies is far from a weak instrument.
[  To test this, we use a panel regression framework to study the time variation in the effect of clan strength on bank development, with the following specification: We estimate this regression using five-year windows between 1891 and 1935. We interact Clans i with a full set of five-year-window dummies between 1896 and 1935 to examine whether the negative effect of clans was mitigated by the modernisation progress. We omit the 1891-1895 period. Thus, all estimated coefficients are calculated relative to this baseline period, which is before the emergence of modern banks. To account for the changing effects of prefectural geographic correlates on banks over time, we also interact our set of geographic controls from Table 1 with the five-yearwindow dummies. Pref i and Period t capture prefectural and five-year fixed-effects, respectively. [ Figure 5 about here]

Foreign and Official Banks as a Placebo
To provide further support for the claim that the effect of clans on banking does not reflect unobserved local factors (e.g., business tradition or strength of other social organisations), we conduct a placebo test by regressing a prefecture's density of foreign and official banks on its genealogy density. In the early years of the expansion of foreign banks in China, the process was primarily driven by Western powers to support their foreign businesses rather than to serve the local population (and thus these banks did regression results show that clan density has no effect on foreign and official bank density (columns 1-3, Table 4). A remaining concern is that the spatial distribution of foreign and official banks is quite minimal when compared with that of private modern banks, with the former located mainly in the big cities. To rule out the possibility that the insignificant clan effect on foreign and official banks is driven by their small regional variation, we restrict the analysis to big cities, i.e., the 75 prefectures with an urbanisation rate in 1910 above the sample mean of 6.76% (columns 4 and 5). This does not change the result based on the whole sample; the clan still has no effect on foreign and official banks. [

Confucian Clans or Conservatism?
Given the extreme emphasis of Confucianism on the status quo and stability -as it is conservative, averse to change and learning from outsiders, and hostile to commerce (Weber, 1905;Landes, 2006) -we expect that in prefectures with strong clans, people had little incentive or desire to start modern banks or any new ventures. For this reason, the negative clan effect that we demonstrate might be driven by Confucian conservatism towards new financial institutions (in which case Confucianism is still not a friend of modern finance), rather than by substitutive competition between formal finance and the Confucian clan.
To rule out this channel, we conduct two separate exercises. First, we examine the effect of clans on each prefecture's number of traditional Chinese banks. There were two main types of traditional banks in historical China. The first were money houses (qian zhuang) or native banks that arose from convenience shops, providing currency exchange and small loan services to petty merchants and households. The Huang (2002) provides systematic records of the prefectural distribution of Shanxi banks and their branches spanning from the early 19 th century to the early 20 th century.
We use the annual average number of traditional banks in each prefecture between circa 1820 and 1912 and construct a measure of traditional bank density in the same way as for the other bank density measures. Again, clan strength has a significantly negative impact on a prefecture's traditional bank density (columns 6-8, Table 4). The instrumented results indicate that doubling the clan density reduces the number of money houses by approximately 30%. 28 28 A related concern is whether the negative effect of clans on modern banks derives from the weak presence of traditional banks. To test this possibility, in a separate exercise, we control for the traditional bank density in the same set of regressions as in Tables 1 and 2. In addition, given the possible effect of foreign and official banks on the development of private modern banks, we also control for the density of foreign and official banks. The results show that the clan effect on modern banks remains robust (see Online Appendix Table A2).
To further ensure that our result is mainly driven by the Confucian clan rather than conservatism, we use the number of Confucian temples to proxy for overall Confucian cultural influence in each prefecture and contrast it with the clan measure in terms of impact on modern bank density. Confucian temples were built to worship Confucian sages and local eminent Confucian exemplars, and hence their number serves to gauge the comprehensive influence of Confucian culture (not just the clan as an economic institution) in a region (Kung and Ma, 2014). Confucian temples were mainly the product of government initiatives to strengthen the ruling culture, whereas clan formation and organising were private initiatives. Specifically, we use the log number of Confucian temples per 10,000 members of the population in the 19 th century, with temple data obtained from the provincial gazetteers compiled at various time points throughout that century. Online Appendix Figure A3 depicts the prefectural distribution of the number of Confucian temples. The results in Table 5 show that the negative clan effect on banking remains robust after controlling for the density of Confucian temples. Confucian temples have a negative effect on modern bank density, but with only a marginally statistically significant coefficient, which is much smaller than that of the clans. Therefore, Confucian conservatism did dampen incentives to create banks, but the Confucian clans as an internal financial market played a much more important role in inhibiting modern finance.

Clans and Weak Demand for External Finance
We argue that the negative impact of the clan on financial development is primarily due to the suppressed demand for external finance. Ideally, we would analyse data on each prefecture's aggregate financial demand, total intra-clan financial transaction volume and total external finance volume. However, such detailed data are not readily available even today, let alone for the early 20 th century. It is thus difficult to conduct a direct breakdown analysis of intra-clan finance and external finance. We focus on the regional interest rates during 1912-1936 because the records during this period are most complete and suitable for inter-regional comparison.
Another reason is that this period saw the genesis of China's modern financial markets. and 1936, ranging from 0 to nearly 100% (the mean is 32%). Note that the interest rates in prefectures with more clans were significantly lower than in other areas.
[ Figure 6 about here] We formally examine the effect of the Confucian clan on interest rates in Table   6, where observations with interest rates above 100% are removed and clan density is and 5) bank (private, official and foreign banks). In addition to the controls used in the earlier tables to capture prefectural characteristics, we control for the term to maturity (one month to over three years) of each loan, and whether the loan occurred in a rural area. We control for provincial fixed effects because interest rates were affected by local currency and inflation factors, which mainly varied due to provincial level factors in early 20 th century China (Dai 1993). To adjust for the temporal change in inflation, we also include the year-province fixed-effects.
Our log-on-log regression results show that interest rates were significantly lower in prefectures with higher clan density, supporting our main hypothesis. These results are robust across specifications with different controls and when using distance from [ Table 6 about here] Thus, intra-clan financing and resource pooling must have been more prevalent in regions with strong clans. Although there are no systematic statistics on intra-clan pooling and sharing services, case studies of powerful clans document practices such as land trusts and alms relief (Section 1), in which the clans usually provided low-or zero-interest loans to members (Tang, 1997). In the dataset used here, the average interest rate for intra-clan loans (23%) was also much lower than that for the other loans (32%). It is possible that the average risk level of loans in prefectures with strong clans was lower than in other prefectures, making the equilibrium interest rates in the former prefectures lower. Unfortunately, we do not have loan-or region-specific risk data for each historical time period; however, in addition to the loan-specific controls described above, we control for each prefecture's geographic and economic prosperity indicators in Table 6. This should correct for each prefecture's socioeconomic characteristics, such as business risk and general economic risk. Furthermore, it is possible that regions with strong clans were also better financially integrated with other regions, in which case their financial supply was higher and the average interest rate lower, and not due to the clans' resource pooling services. This channel is ruled out by the inclusion of prefecture-specific controls such as distance to treaty ports, rivers, the coast, the provincial capital, telegraph stations and postal offices.
We informally assess the extent of intra-clan finance by checking the number of clan-based lending records in the China Historical Interest Rate database. As illustrated by the 2010 survey data (discussed below), intra-clan borrowing rarely took the form of a written contract. Rather, it was performed orally, leaving no written record. However, borrowing between non-relatives (especially borrowing from financial institutions) was more likely to be recorded. Thus, there may be a strong bias against intra-clan financial transactions in the surviving records on private finance. Of the 14,309 loan transactions in the database between 1912 and 1936, 483 (3.4%) were 29 A remaining concern is that the lower interest rates in the strong-clan regions may be due to the high ratio of productive to emergency consumption loans. A possible explanation is that prefectures dominated by clans may have had less demand for emergency loans, which usually have high interest rates. We use the share of loans borrowed by commoners (out of those borrowed by merchants, banks and other business organisations) as a proxy for the share of emergency loans and regress it on clan density, with the same set of controls as reported in Table 1. Clans do not significantly reduce the number of local emergency loans (Online Appendix Table A3). This result may be due to the lack of information on the specific use/situation of each loan.
provided by clan organisations. We call these 'clan loans'. To account for possible recording biases, we calculate the ratio of the number of clan loans to the total number of loans in each prefecture. To mitigate the effect of outliers, we take the natural logarithm of 1 plus this value in percentage points. Figure 6(b) shows that most clan loans in the database were concentrated in regions with high clan density, suggesting a positive correlation between clan strength and intra-clan lending. The regression results in Table 7 confirm that the share of clan loans was significantly higher in prefectures with strong clans, implying that the clans indeed played an important role in providing informal financial services for members. This result remains robust to the inclusion of prefectural characteristics and other specifications.

Long-term Effects of Confucianism on Finance
Despite China's modernisation effort during the early 20 th century, Confucianism continued to stifle financial development by reducing the demand for external finance.
Does China's prefectural clan culture still shape demand for and attitudes towards external financial markets? As Confucianism developed over more than two millennia, its effect on Chinese people's way of life is likely to have persisted. As Greif and Tabellini (2017) point out using contemporary survey data, clan culture and clan-based cooperation quickly revived after the market reforms that started in 1978, although Confucianism was denounced and clan structures were destroyed before and during the Cultural Revolution (1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975)(1976)) (see also Peng, 2004;Hu and Tian, 2018;and our Online Appendix Figure A4). However, we seek to determine whether the negative effect of the clan on financial development has survived to the present day. As seen in the next section, the Confucian clan continues to hamper external finance in modern China.
For this exercise, we measure each prefecture's clan strength by including all genealogy books compiled prior to 2009 that are kept in the Shanghai Library's (2009) Comprehensive Catalogue on the Chinese Genealogy. As many genealogy books were written long ago and are unlikely to reflect the robustness of clans in China today, we also use the number of genealogy books compiled during the recent reform period  as an alternative measure of clan strength. We normalise the number of genealogy books by population size in 2009.

28
To test whether past clan strength predicts today's clan culture, we rely on two trust variables based on data from the CGSS 2010. 30 We first collect the answers to the following questions: 1) 'To what extent do you trust your family members and relatives?' and 2) 'To what extent do you trust other people (outside the family)?' Answers to both are graded on a 1-5 scale, where 5 denotes complete trust and 1 denotes complete distrust. For each of the two questions, we convert the graded values to a dummy variable: 1-3 are coded as 0 (less trust) and 4-5 as 1 (more trust).
Next, we regress the two dummy variables of trust on clan density at the prefectural level by controlling for individual characteristics that may affect trust, such as education, age, gender, rural-household dummy, ethnic-minority dummy and number of kin; for prefectural factors such as GDP per capita in 2010, jinshi degree holders , dummy for Protestantism in 1920; and for the set of geographic factors reported in Table 1. We report the Probit estimates in Table 8, where we instrument clan density by using distance to the Zhu Xi academies. The results show that in prefectures with historically stronger clans, people report greater trust in relatives (columns 1 and 2) and distrust outsiders more (columns 4 and 5). Note that when historical clan density is replaced by modern clan density , the negative clan effect is even stronger. This is true for the other specifications in Tables   8 and 9. Thus, a region's clan culture from the past continues to determine clan ties today.
To gauge the influence of past Confucian clan culture on members' business conduct in the present, we use the answers to another CGSS question: 'To what extent do you prefer to use formal contracts when doing business with relatives?' (1-5 scale).
Likewise, we convert each answer into a dummy variable: 1-3 are coded as 0 (less likely to use formal contracts) and 4-5 as 1 (more likely). The instrumented results illustrate that people in prefectures with historically stronger clans do not prefer to use formal contracts in business dealings with relatives (columns 7-9). In such an environment, the demand for, and hence the supply of, formal impersonal institutions, legal and otherwise, cannot be high.
[ Table 8 about here] A probable direct consequence of the persistence of clan culture is that people from stronger clans are less likely to use formal external finance. To test this, we proxy for individual participation in formal finance based on the answers to the following 30 The CGSS, conducted by Renmin University of China, is the most representative social survey in China. In 2010, 11,783 people were interviewed across all provinces.

29
CGSS question: 'Do you borrow money from banks (including mortgage loans)?' The results are as expected: people in prefectures with historically stronger Confucian clans are less likely to take out bank loans (Table 9, columns 1-3). Moreover, we find that even when people do borrow from a bank, the typical loan size is significantly smaller in prefectures with higher clan density in the past (columns 4-6). These results are robust to the other specifications used in earlier exercises.
[ Table 9 about here] Last, we find that in prefectures with a stronger clan culture, the level of financial development is also lower today. To proxy for financial development, we use both 1) the logarithm of total bank loans and deposits per 10,000 members of the population, and 2) the logarithm of bank loans and deposits to GDP, based on data obtained from the 2010 City Statistical Yearbook of China. In addition to the controls used in Table   1, we control for the density of banks in 1936, jinshi degree holders  and Protestantism in 1920 to capture their long-term effects, along with contemporary economic prosperity, measured by GDP per capita in 2010. 31 Columns 7-12 of Table   9 show that there is less banking per capita and that banking is a smaller part of the local economy in prefectures that had stronger clans in the past or have higher genealogy density today.

Conclusion
Based on data from both 1897-1936 and recent decades, our empirical work shows that since modern finance was introduced to China in the late 19 th century, it has faced subtle but significant competition from the traditional resource pooling and risk sharing services provided by the Confucian clan. After more than two millennia of efforts to perfect the clan by Confucian scholars and imperial courts, the clans served sufficiently well as internal financial markets and business corporations. It is then no surprise that many aspects of modern finance were, and may still be, redundant for many ordinary people and businesses. Therefore, when formal finance was first introduced, few individuals had an incentive to switch to it, limiting the demand for financial development. In addition, it was not easy to stimulate the supply side of arms-length finance. Since the Han dynasty, and especially since the Song dynasty, the Confucian clan has provided much of the solution and hence hampered the development of impersonal market institutions. Although China had introduced money (Goetzmann 31 GDP per capita is not included in examining the clan effect on bank loans and deposits to GDP.

30
and Koll, 2005) and simple lending-borrowing practices prior to Confucius' time, 32 not only did external finance fail to develop endogenously, but when it was eventually transplanted, society also remained comparatively uninterested. Even in today's China, regions that have traditionally had strong clans still maintain a relatively low demand for modern finance, relying instead on intra-clan services.
A key insight from our analysis is that financial development is path dependent.
This contrasts with the focus of financial development studies on the institutions that are common in well-developed financial markets. Our study shows that to understand why some societies have developed the necessary institutions for finance whereas others have not, we need to go back in history. Some societies may have developed alternative systems with the same resource pooling and risk sharing roles as formal finance; therefore, these societies have not perceived the need to introduce or entrench the kinds of impersonal institutions needed for financial development. As discussed in Greif and Tabellini (2017) (Goetzmann, 2005). . 'Weak clan region' refers to the prefectures whose number of genealogy books per 10,000 people in 1896 is below the sample mean (0.36) of all 284 prefectures, whereas prefectures above this mean are defined as strong clan regions. We exclude 36 prefectures that were treaty ports or had an outlier number of clans (top 5% tail) by 1897. To mitigate geographic and economic heterogeneities, as shown in Figure 2(b), we further exclude the five frontier provinces in western China (Guizhou, Yunnan, Guangxi, Sichuan and Gansu).  02 1900 1905 1910 1915 1920 1925 1930 1935 Weak clan region Strong clan region  03 1900 1905 1910 1915 1920 1925 1930 1935 Weak clan region   Table 1. We randomly select three counterfactual Zhu Xi academies from two different samples: one sample consists of the 54 prefectures with academies during the Zhu Xi era (1130 to 1200) (Figures (a), (c) and (e)), whereas the other sample consists of the 90 prefectures whose density of jinshi degree holders was above the national median in the Song dynasty ((b), (d) and (f)). Both samples exclude the three prefectures with actual Zhu Xi academies and their adjacent prefectures. In each sample, we repeat the drawing 2,000 times and thus construct 2,000 sets of counterfactual academies. The p-value refers to the share of the number of counterfactual coefficients that are greater than the coefficient of distance to the actual Zhu Xi academies (red line). For each figure, the vertical axis stands for the number of occurrences for the regression coefficient lying in a given range on the horizontal axis.  1900 1905 1910 1915 1920 1925 1930 1935 - 1900 1905 1910 1915 1920 1925 1930 1935 41 (a) (b) Figure 6. Distribution of Interest Rates andIntra-Clan Lending, 1912-1936 Notes: The data of interest rates and clan lending are obtained from the China Historical Interest Rate database constructed by Chen et al. (2016). The three dark points refer to the locations of the Zhu Xi academies in the 12 th century. In (b), clan lending refers to percentage of credit provided by clan organisations. Notes: The dependent variable is the log number of Chinese modern banks in each prefecture in 1936 in columns 1-6, and the log annual average of the number of Chinese modern banks in each prefecture between 1897 and 1936 in column 7. We normalise the number of banks by population in 1920 in columns 1-6 and by the average population of 1910 and 1920 in column 7 (in units of 10,000). Clan is measured by the number of genealogy books that had been compiled by the year 1896 in each prefecture. We normalise the number of genealogy books by the average population between 1393 and 1851 (in unit of 10,000) and take the logarithm. Geographic controls include distance to treaty ports in 1896, distance to river, distance to coast, distance to province capital and an eastern China dummy. Robust standard errors are given in parentheses. Standard errors in brackets are clustered within a radius of 136 kilometres to account for possible spatial correlation based on Colella et al. (2019). ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.  Notes: The dependent variable is the log number of Chinese modern banks in each prefecture in 1936 in columns 1-3, and the log annual average of the number of Chinese modern banks in each prefecture between 1897 and 1936 in column 4. We normalize the number of banks by population in 1920 in columns 1-3 and by the average population of 1910 and 1920 in column 4 (in unit of 10,000). Clan is instrumented by the log distance to the Zhu Xi academies. Geographic controls include distance to treaty ports in 1896, distance to river, distance to coast, province capital and an eastern China dummy. Additional controls include industrial firms, commercialisation, urbanisation rate, telegraph, post offices, Protestantism, jinshi degree holders, and civil conflicts. Robust standard errors are given in parentheses. Standard errors in brackets are clustered within a radius of 136 kilometres to account for possible spatial correlation based on Colella et al. (2019). ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Notes: The dependent variable is the log annual average of the number of foreign and official banks (per 10,000 population) in each prefecture between 1903 and 1936 in columns 1-5. Columns 4 and 5 restricts the sample to the prefectures with urbanisation rates in 1910 above the sample mean of 6.76%. Dependent variable in columns 6-8 is the log number of traditional Chinese banks, qianzhuang (money houses) and piaohao (Shanxi banks), which were established in each prefecture in 1820 to 1912. We normalise it by population (per 10,000 people) of 1910. Clan is instrumented by the log distance to the Zhu Xi academies. Controls in columns 1-5 are the same as those of column 2 of Table 3. Controls in columns 6-8 are the same as those of column 1 of Table 3. Robust standard errors are in parentheses. Standard errors in brackets are clustered within a radius of 136 kilometres to account for possible spatial correlation based on Colella et al. (2019). ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.  Notes: The dependent variable is the logarithm of the interest rate of each lending record between 1912 and 1936. Clan is measured by the log number of genealogy books per 10,000 people in 1910. In columns 5-7 clan is instrumented by the distance to the Zhu Xi academies. Lending controls include maturity, rural area dummy, and lender category dummies (individual, community, merchants, traditional banks and modern banks). Geographic controls and additional controls are the same as those in Table 3. Robust standard errors adjusted for clustering at the prefectural level are given in parentheses. Standard errors in brackets are clustered within a radius of 136 kilometres to account for possible spatial correlation based on Colella et al. (2019). ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Notes: The dependent variable is the logarithm of the share (percentage points) of intra-clan lending in all lending records at the prefectural level between 1912 and 1936. Clan lending refers to lending transactions by clan organisations to members. Clan and controls are the same as those of Table 3. In columns 5-7, clan is instrumented by the distance to the Zhu Xi academies. Robust standard errors are given in parentheses. Standard errors in brackets are clustered within a radius of 136 kilometres to account for possible spatial correlation based on Colella et al. (2019). ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.  , Protestantism in 1920, and GDP per capita in 2010. Robust standard errors adjusted for clustering at the prefectural level are given in parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.   Table 8, and both are instrumented by the distance to the Zhu Xi academies. Controls are the same as Table 8, except that the log number of banks in 1936 (per 10,000 people) is also included. Robust standard errors adjusted for clustering at the prefectural level are given in parentheses.

Figures and Tables
Standard errors in brackets are clustered within a radius of 136 kilometres to account for possible spatial correlation based on Colella et al. (2019). ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Figure