This study examines the agency costs of 314 family firms listed on Shanghai and Shenzhen Stock Exchange in China. The results show that agency costs of family firms increase significantly with the enhancement of familys ownership and the separation of ownership and control, while family involvement in management can reduce the agency costs. By including the variable proxying internal control and the interaction term between internal control and family involvement in the empirical model, we still find that the effects of family involvement on agency costs are contingent on the quality of internal control.
Family firms ; Agency costs ; Internal control ; Confucianism
Family firms are the most ancient form of organizational structure, and play an important role in the modern economic life. Approximately two-thirds of private businesses are family owned over the world (Neubauer and Lank, 1998 ). In the Fortune global 500 companies, 175 firms are controlled by family. Since the reform and opening up, Chinas family firms have experienced unprecedented development. According to the latest statistics released by the State Administration for Industry and Commerce of China, the number of private enterprise dominated by family has exceeded 10 million by the end of 2014, accounting for more than 60% of Chinas GDP.
Family firms are linked with blood relationship and kinship ties. In light of classical agency theory, such a guan-xi culture is helpful to alleviate the interest conflict within the firm and reduce the agency costs. Therefore, family firms are considered the most efficient forms of organization (Fama and Jensen, 1983a and Daily and Dollinger, 1992 ). However, Schulze et al., 2001 and Schulze et al., 2003 challenge the classical agency theory. They argue that family firms may be subjected to higher agency costs because of conflicts associated with private ownership and owners self-control, as well as those posed by asymmetric altruism, so that the performance is rather lower than that of non-family firms.
This study contributes to the literatures on agency problems within family firms in the following ways. Firstly, most researches treat family firms as a homogeneous group, ignoring the difference in the way that the controlling family may influence a firm. Following the approach of Villalonga and Amit (2006, 2009) , this study takes into account the differences in family firms by distinguishing among three fundamental elements in the family involvement, namely, ownership, control, and management. Secondly, in a social context in which family takes on particular meaning, we are able to investigate the influence of traditional Confucian values on the agency costs in Chinese family firms. The unique sample not only fills the empirical void for emerging markets but also enriches the understanding of the sector. Finally, considering the characteristics of Chinas Accounting Standard and the availability of data, we use perquisite consumption and inefficient investment as two new proxies for the agency costs.
The remainder of the paper is organized as follows: In the next section we present the theoretical analysis and develop our hypotheses. We then detail the methodology in the third section. We present the empirical results and discuss the implication of our findings in the fourth section. Finally, we conclude the paper with its limitations and provide some suggestions for future research.
Interest conflicts and information asymmetry are the basic causes of the agency costs (Jensen and Meckling, 1976 ). The classical agency theory documents that the family firms represent one of the least costly and most efficient forms of organization for the following reasons. First, in the family business, ownership mainly concentrates in the family members with blood relationship, such a “private property” can ensure the consistency of family agents’ pursuit with the interests of the whole enterprise, so as to alleviate the agency tension caused by interest conflict (Fama and Jensen, 1983a ); Second, the mutual communication and coordination among family members are more convenient, which help to reduce the agency costs elicited from information asymmetry (Fama and Jensen, 1983b and Daily and Dollinger, 1992 ). Accordingly, the following hypothesis is proposed:
The agency costs of family firms are lower than non-family firms.
However, family firms are not a homogeneous group with the same sets. Family involvement always differs greatly in ownership, control and management. The controlling family always holds a larger proportion of equity, which can ensure the consistence of interests inside enterprise and reduce the agency costs caused by interest conflicts. But with the expansion of scale, family firms tend to attract external capital. Through the cross-shareholding and “pyramid” structure, controlling shareholders can achieve dominant position with less ownership. Such a separation of ownership and control is most pronounced among family-controlled firms (Claessens et al., 2000 and Faccio and Lang, 2002 ). The entry of external capital exacerbates the interest conflicts between the controlling family and other shareholders and exerts adverse effects on the control of agency costs. In addition, family involvement in management is also a key attribute distinguishing family firms from non-family firms. Compared with the non-family managers, family managers tend to view the fulfilment of family business goals as intrinsic. They always subordinate their personal interests to working towards the organizational interests, leading to no or little agency problems (Chrisman et al., 2007 and Karra et al., 2006 ). Accordingly, the following hypotheses are proposed:
Family ownership relates with the agency costs of family firms negatively.
The separation of family ownership and control relates with the agency costs of family firms positively.
Family involvement in management relates with the agency costs of family firms negatively.
While family involvement is common practice in family firms across cultures, its meaning and role is contextual and culture-specific. China is the birthplace of the Confucianism. Confucianism is woven into the very fabric of Chinese society, and contributes distinctive characteristics to Chinese family firms. Confucianism stresses the reciprocal benevolence between persons. Such a reciprocal relationship underpins the family hierarchical order and social harmony in China. In the business context, the benevolence of family business owners towards other members always manifests in the positioning them in senior managerial roles. To return the favour, family managers always display strong psychological allegiance and commitment to the enterprise. Mutual obligation can reduce individual opportunism of family agents’ and agency costs of firms accordingly (Kim and Gao, 2013 ). More importantly, Confucianism promotes filial piety, submission and obedience of subordinates towards superiors, which are likely to induce steward-like behaviour of family manager and contribute to the solution of agency problems (Bell, 2010 ).
However, Confucian values prevalent in Chinese family firms may also exert negative effects on the agency problems. As mentioned above, the personnel recruitment and promotion of key positions are often made based on family superiors personal preference, rather than formal systems and procedures. It easily leads to the entry of unqualified family members into the business. Even worse are that these people still enjoy the privileges and their behaviours are always beyond the control. In addition, Confucianism stresses the strict ethics and hierarchical order, which easily induce the paternalism of family superior. Owning to the lack of scientific decision-making procedures in many Chinas family firms, the irrational strategies and inefficient investment decisions are often made by family superior based on individual experiences or judgments, thus exacerbating the agency problems. This is the important reason that many Chinese family firms fail in operation.
Fortunately, with the entry of western management philosophy and the promotion of modern enterprise system by Chinese government, nepotism and paternalism permeating in Chinese family firms have been inhibited to a large extent in recent years. Most Chinese family firms have established formal internal control system to offset the negative influence of Confucian values. Internal control is a process which is carried out by board of directors, management and other staff and aims to provide reasonable assurance for the following objectives: (i) the efficiency and effectiveness of operation; (ii) the reliability of financial reporting; (iii) compliance with related laws (COSO, 1992 ). In light of contingency theory, the behaviour of agents is contingent on situational or psychological factors specific to a firm (Davis et al., 1997 ). Under the pressure and restraint of internal control, the irrational and opportunistic behaviour of family members can be prevented and discovered in time. More importantly, internal control can ensure the reliability of financial reports (Ashbaugh-Skaife and Collins, 2008 and Chan et al., 2008 ), which facilitates principals to obtain relevant information about agents’ behaviour and firms performance, so as to reduce the agency costs caused by information asymmetry. Accordingly, the following hypothesis is proposed:
The association between family involvement and agency costs is more pronounced in firms with high-quality internal control.
We start with a base sample consisting of all the non-financial firms listed on the Chinese Shanghai and Shenzhen Stock Exchange in 2012. We then identify family firms using the following criteria: (i) The ultimate controlling shareholder of a firm is individual or family; (ii) The controlling family is the largest shareholder; (iii) The family ultimate ownership is over 20%. The ownership data are hand collected from firms’ annual reports. We also search relevant information on Internet in order to identify the family members and their relationship. Financial data and corporate governance data are from CCERDATA and http://www.cninfo.com.cn/ . Our final sample consists of 314 family firms.
The variables in the model (1) are defined as following: Invt is the ratio of cash paid for fixed assets, intangible assets and other long-term assets to total assets for year t ; Growtht − 1 is the growth rate of gross operating income for year t − 1; Levt − 1 is the asset-liability ratio for year t − 1; Casht − 1 is the ratio of sum of cash and tradable financial assets to total assets for year t − 1; Sizet − 1 is the natural logarithm of total assets for year t − 1; Rett − 1 is annual rate of stock-market returns for year t − 1; Invt − 1 is the ratio of cash paid for fixed assets, intangible assets and other long-term assets to total assets for year t − 1; Industry1-11 are dummy variables for industry.
Family firms (FF): We first employ a dummy variable FF to indicate family firms. If a firm is identified as a family firm, FF takes the value of 1; otherwise, FF equals 0.
Separation of family ownership and control (Sep): Following Claessens et al. (2000) , we define Sep as the ratio of FO and FC, where FO is the fraction of cash-flow rights held by family, and FC is the sum of cash-flow rights of the weakest links in the ownership chain.1
Family management (FM): the proportion of family members on the board of directors.
Family involvement (FI): We construct this indicator to capture the comprehensive effect of FO, FC and FM. FI is calculated as the mean of FO, FC and FM.
The quality of internal control (ICI): We use the evaluation method of internal control put forward by Zhang et al. (2011) and employ the assessment scores of the realization level of control objectives as the proxy of internal control quality.3.2.3 Control variables.
Firm size (Size). We use the natural logarithm of total assets as the proxy of firm size. There are two opposite views about the relationship between the firm size and agency costs. Danes et al. (2007) argue that larger firms usually have sounder systems and richer resources, which helps to reduce agency costs. However, Chu (2009) finds that more intimate relationship of family members are always present in the smaller family businesses, which can alleviate the interest conflict and reduce the agency costs. So the sign of the coefficient of Size is an empirical issue. Firm age (Age). In general, the longer the firms age is, the sounder the governance mechanism is, so the agency costs tend to be lower (Chu, 2009) . But with the advent of new nuclear family, interest conflicts between family members become more and more serious (Karra et al., 2006 ). Therefore, we include the age of firm as the control variable in the model.
Free cash flow (FCF). According to Free Cash Flow Hypothesis (Jensen, 1986 ), executives are more likely to invest and consume more money than needed in the firms with abundant free cash flow and low growth, which leads to the increase of agency costs. So we control for free cash flow in the model and use the following formula to calculate the free cash flow (Jiang and Huang, 2009 ):
FCF = (net cash flow from operating activities − net working capital changes − cash paid for fixed assets, intangible assets and other long-term assets)/total assets.
Debt financing (Lev). Debt contracts not only reduce free cash flow, but also increase the likelihood of bankruptcy of enterprise and unemployment of manager, which is favourable to inhibit the opportunistic behaviour of agents and reduce the agency costs (Jesen, 1986). So we use the debt-to-assets ratio to measure the level of debt financing.
Independent audit (Big4). High-quality audit of Certified Public Accountants is proved to be an effective governance mechanism to mitigate the agency costs (Watts and Zimmerman, 1986 and Beatty, 1989 ). Prior empirical results show that the audit quality of Big-4 firms is higher than that of Chinas domestic accounting firms (Qi et al., 2004 and Pan, 2008 ). So we control for the type of accounting firms by using an indicator variable Big4 , which equals to one when the auditor is a Big-4 firm; and zero otherwise.
Industry (Indus1-11 ). According to the industry classifications set by China Securities Regulatory Commission, we include 11 control variables of industry in the model.
According to the hypotheses developed and the variables defined above, the following multiple linear regression models are employed to analyze the agency costs of family firms: the model (2) only include the proxy of firm property (FF) and control variables to test H1; the model (3) includes family ownership (FO), the separation of family ownership and control (Sep) and family management (FM) to test H2-42 ; In the model (4), we introduce family involvement (FI), internal control quality(ICI) and an interaction term between FI and ICI to test H5.
Descriptive statistics for the independent variables and dependent variables are provided in Table 1 . The mean (median) of family management (FM) is 15.9(12.5) and far smaller than that of family ownership (FO) and family control (FC), which indicates that family members’ participation in management does not match with the level of family ownership and control in Chinese family firms. The mean of separation of family ownership and control (Sep) is 0.8395. According to prior studies, the average separation of ownership and control for 2980 corporations in nine East Asian countries is 0.746 (Claessens et al., 2000 ) and that of western family enterprises is 0.868 (Faccio and Lang, 2002 ). Our result indicates that the separation degree of ownership and control of Chinese family firms is still moderate.
The Pearson correlation matrix is provided in Table 2 . Except the correlation between family ownership (FO) and family control (FC), other correlations among the independent variables are all less than 0.5 and exhibit no serious multicollinearity. Dependent variable AC1 is significantly negatively associated with family ownership (FO), family control (FC) and family management (FM), which indicates that firms with higher level of family involvement are subjected to lower agency costs. Such a result is consistent with H2 and H4 . As expected, dependent variables (AC1, AC2) are negatively associated with the quality of internal control (ICI).
In order to test hypothesis H1, we regress model (2) for our base sample of 2177 non-financial companies listed on Chinese Shanghai and Shenzhen Stock Exchange in 2012. The results are reported in Table 3 . The Adjusted R2 equals to 0.1517 (when the dependent variable is AC1) and 0.0814 (when dependent variable for the AC2). The variance inflation factor (VIF) of independent variables are all less than 2 and the average VIF is 1.28, which also indicates that multicollinearity between independent variables is not serious. The coefficient for FF is 0.002(P = 0.375) when dependent variable is AC1, which is inconsistent with the expected sign; the coefficient for FF is −0.027(P = 0.404) when dependent variable is AC2, which is consistent with expected sign but not significant. These results indicate that Chinese family firms do not outperform non-family firms in controlling agency costs, and that the perquisite consumption of senior executives is even higher than that of non-family firms. Our hypothesis (H1 ) is not supported. We interpret the results that it may reflect the complexity of agency problems of family firms. Family firms are not a homogeneous group. The differences in family involvement and governance mechanism may influence the level of agency costs, therefore further analysis is needed.
To further evaluate the effect of family involvement on the agency costs, we confine our sample to the family-controlled firms and introduce the variables of family ownership (FO), the separation of family ownership and control (Sep) and family management (FM) in model (3). The regression results are reported in Table 4 . Some findings are noteworthy. First, compared with model (2), the explanatory power of model (3) is improved significantly. Adjust R2 of model (3) equals to 0.3072 (when the dependent variable is AC1) and 0.1058 (when the dependent variable is AC2), which shows that the inclusion of family involvement does help to explain the agency costs of family firms in China. Second, the coefficient of family ownership (FO) is positive, which is inconsistent with H2 . One possible explanation for this is that many controlling families in China just hold the shares and do not participate in the operation of the firm, so the influence of family ownership (FO) on agency costs is weaken. Third, the coefficient of the separation of family ownership and control (Sep) is significantly negative at the 0.05 level. This result support our hypothesis (H3) that when the ownership of the family is different from (especially less than) the control, family agents are more likely to seek personal gains or family interests, thus exacerbating the agency costs of the whole business. Finally, the coefficient of family management (FM) is significantly negative, which is consistent with H4. Family members being the agents, their purposes are more closely related with that of the whole firm, so that family involvement in management can effectively reduce the agency costs.
In order to evaluate the moderating effect of internal control on the relationship between family involvement and agency costs, we include an interaction variable between family involvement (FI) and internal control (ICI) in model (4). Regression results are shown in Table 5 . When the dependent variable is AC1(AC2), the coefficient of internal control (ICI) is −0.117 (−0.080) respectively, which indicates that internal control not only effectively reduce the perquisite consumption of executives, but also inhibit the inefficient investment. The coefficient of family involvement (FI) is insignificantly negative, but the coefficient of the interaction variable between family involvement (FI) and internal control (ICI) is significantly negative. Such a result found in our study shows that the relationship between family involvement and agency costs is moderated on the quality of internal control. In other words, only firms establish high-quality internal control, can family involvement reduce the agency costs indeed. As a result, our hypothesis (H5) is proved.
Family firms are believed to be the organization with lowest agency costs (Fama and Jensen, 1983a and Fama and Jensen, 1983b ). In this paper, we compare and analyze the agency costs of 314 family firms and 1863 state-controlled enterprises listed on the Chinese Shanghai and Shenzhen Stock Exchange in 2012. The empirical results show that Chinese family firms do not have significant advantage over state-controlled enterprises in controlling agency costs, and the agency costs associated with perquisite consumption are rather higher than that of state-controlled enterprises. We still find that the agency costs of family firms rises significantly with the increase of family ownership and the separation of ownership and control, and family involvement in management is significantly negatively related with the agency costs. At last, by introducing the variable of internal control and interaction variable between family involvement and internal control in the model, we find that the impact of family involvement on agency costs is contingent on the quality of internal control. These findings contribute to the current empirical literature on the agency problem of family firms and provide useful implications for practice.
However, this study has some limitations. Firstly, because most of the data used have to be hand-collected, we only use the cross-sectional data of firms in 2012 as the research sample. If panel data are used in analysis, the results may be more robust and convincing. Secondly, our sample only includes listed family firms, which only account for a small portion of the whole group of family firms in China. Therefore, caution should be exercised in generalizing these results to the family firms of the all. Finally, we only study the Agency Problem I, which exists between the stockholders and managers. As for the Agency Problem I, which exists between majority and minority shareholders, has not been take into account in this study. Further research involving more representative samples and more comprehensive aspects will allow researchers to develop a better understanding of agency problem in family businesses.
The authors declare that there is no conflict of interest.
1. Suppose that a family owns 10% of the stock of Firm A, which in turn has 20% of the stock of Firm B. We would say that the family owns 2% of the cash-flow rights of Firm B (FO is 2%) and the family controls 10% of Firm B (FC is 10%).
2. As FO is highly correlated with CO (correlation is 0.8388, p < 0.001), we do not include them in the model simultaneously to avoid multicolinearity.