Trust, family businesses and financial intermediation

CORFIN-00886; No of Pages 24 Journal of Corporate Finance xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Journal of Corporate Finance journal homepage: www.elsevier.com/locate/jcorpfin

Trust, family businesses and financial intermediation☆ Massimiliano Stacchini ⁎, Petra Degasperi Bank of Italy

a r t i c l e

i n f o

Available online xxxx JEL codes: G21 Z13 Keywords: Banking Family firms Social capital

a b s t r a c t This paper analyzes whether interpersonal trust affects the agency costs of family-controlled firms' debt. Our results are threefold. First, we find that banks apply a discount to the interest rates charged to family firms, whose size decreases considerably for contracts stipulated in high-trust areas. Second, as a response to the (unexpected) liquidity shock affecting the interbank market in August 2007, banks further increased the discount associated with family control. Third, we have no evidence of lender-corruption effects since the ex-post performance of the (cheaper) loans extended to the family firms is superior to that of their peers. These results suggest that banks deem the incentive structures prevailing in family firms to be able to attenuate the higher risks of expropriation run by lenders in areas where agency conflicts are greater, due to lack of interpersonal trust. Our findings are robust to the self-selection and omitted local variable problems as well as to credit demand-side effects. We model family control as an endogenous choice, introduce local-level fixed effects and use the heterogeneous exposure of Italian banks to the 2007–09 financial crisis – and the fact that Italian firms have more than one lender – to fully absorb changes in credit demand schedules. © 2015 Elsevier B.V. All rights reserved.

1. Introduction The literature on the link between family businesses and financial intermediation is scant and does not reach clear conclusions (Anderson et al., 2003; Bennedsen et al., 2010; La Porta et al., 1997; Villalonga and Amit, 2006). According to some authors, the incentives of family block-holders and their lending banks tend to be more closely aligned than those of lenders and other firms' large shareholders – such as banks, investment funds and widely held companies – and result in lower monitoring costs for banks. Banks evaluate family firms as borrowers having lower repayment problems, due to their stable and long-term oriented business model, which arises from the connections existing between the firm prospects and the family prosperity, the desire of family owners to preserve the family's reputation and to transmit the firm's wealth to descendants (Anderson et al., 2003). By contrast, other scholars point out that large family shareholders may have an incentive and the power to expropriate minority shareholders by favoring family members in the selection of managers or imposing altruistic values in the conduct of business, i.e. by following practices that reduce efficiency and the ability to service debt (Caselli and Gennaioli, 2013; Pinheiro and Yung, 2014;

☆ The views expressed here are those of the authors and do not necessarily represent those of the institution with which they are affiliated. We are grateful to Morten Bennedsen, Joseph Fan, Yishay P. Yafeh (the discussant) and the participants at the Journal of Corporate Finance Special Issue Conference on Family Firm Governance, Beijing, China, June 2013. This research has also benefited from the helpful suggestions from Matteo Bugamelli, Riccardo De Bonis, Valeria Ferroni, Francesco Manaresi, Matteo Piazza and Robert Waldmann. We also thank Danilo Liberati for the helpful data collection. ⁎ Corresponding author. E-mail address: [email protected] (M. Stacchini).

http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006 0929-1199/© 2015 Elsevier B.V. All rights reserved.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

2

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Villalonga and Amit, 2006). Moreover, Ellul et al. (2007) argue that a family block-holder's power to extract private benefits of control depends on the enforceability of contacts in the country where the firm operates. This study empirically analyzes the link between family businesses and financial intermediation, by focusing on interpersonal trust1 and its heterogeneous distribution across Italian regions, as the key factor for discriminating between the two competing views. The role played by trust in financial intermediation is well known in the literature. Whether individuals are willing to sign financial contracts as well as their propensity to buy stock depends not only on the enforceability of contracts, but also on the extent to which they trust the counterparty (Guiso et al., 2004). We test the idea that the value of family businesses in reducing the agency costs of debt is higher in low-trust environments, i.e. where individuals are more inclined to behave opportunistically – as the tendency to cooperate is weaker – and borrowers are more prone to invest in risky projects without bearing the costs of downside failure. We verify the implications of Putnam (1993) and Fukuyama's (1995) thesis that where cooperation mechanisms are weaker due to the lack of social capital, delegation problems and other agency conflicts tend to be mitigated by personal relationships, such as those existing within family members. As for firms located in low-trust areas, our findings show that family controlled borrowers benefit from a loan interest-rate discount compared with non-family firms, suggesting that banks deem the incentive structures of family firms to be less likely to cause expropriation of lenders. By contrast, in trust-intensive areas, family firms lose the discount advantage with respect to their peers, suggesting that where opportunistic behavior and delegation problems are less likely to occur, banks reduce the importance they attribute to family firms when they evaluate borrower risk. We also show that the (unexpected) shock caused on the interbank market by the failure of Lehman Brothers increased the discount that banks applied to family firms operating in low-trust areas. These results are in line with the existing literature showing that family firms suffered relatively less than their peers from the credit tightening that occurred during the 2007–09 financial crisis (D'Aurizio et al., in this issue). We contribute to this strand of the literature by showing that (mainly) in low-trust areas, lenders perceive the rise of bankruptcy risks for family firms to be less intense than for similar firms not controlled by families. When economic conditions worsen, the temptation to respond to business difficulties by engaging in increasingly risky activities (gambling for resurrection) might be stronger where opportunistic behavior is more likely to occur and less intense for borrowers having higher reputation concerns. Furthermore this paper does not find evidence of corruption effects on loan interest-rate discounts granted to family-owned firms by their lenders. Econometric tests show that the lower rates applied ex-ante on loans to family firms are justified ex-post by the objective creditworthiness of these borrowers, highlighted by their loans' better performance. Three factors make Italy an ideal environment for this study. First, it is a paradigmatic case of asymmetric distribution of trust within a country (Banfield, 1958; Putnam, 1993). Such asymmetry, which is one of the drivers of the heterogeneity in the development of the Italian regions, provides the source of variability needed in the data to investigate the trust effects. Second, the significant presence of family firms in the Italian economy – they account for almost 80% of the business – makes it worth studying the financial conditions of these enterprises. Third, the narrowness of the Italian bond and equity markets makes Italian firms highly dependent on bank debt for their external finance; which makes the analysis of loan pricing highly representative of the agency costs of debt sustained by entrepreneurs in Italy. The results presented in this paper are based on a unique loan-level longitudinal dataset made up of more than 107,000 financial contracts signed or renewed in Italy from 2005Q1 to 2011Q4. The performance of these loans has been observed until 2013Q3. Our empirical strategy tackles several issues potentially biasing our estimates: i) the omission of variables that are important at a local level and that might overestimate the role played by local trust; ii) the possibility that the variable indicating whether a firm is owned by a family is endogenous to the firm's prospective earnings and credit conditions; iii) the possibility of the credit demand of family and non-family firms changing over time and confounding the supply-side effects of the shock caused by the failure of Lehman Brothers; and iv) the fact that the explicit components of loan contracts, i.e. the interest rates, are likely to be determined jointly with the implicit ones, i.e. the posting of collateral, so that simultaneity bias may affect loan interest-rate regressions. The first problem is tackled by an empirical strategy exploiting a result of the literature on the determinants of trust. Being social capital a resource of individuals that emerges from social ties, the source of this capital lies with the people (and the environment) a person is related to (Guiso et al., 2004, 2008). The level of social capital prevailing in the area where an individual live affects its propensity to cooperate and generates most of the subjective trusting attitudes needed for financial transactions (Guiso et al., 2004). We move from these arguments to derive our empirical strategy. We focus on contracts stipulated between a bank and a firm located in non-coincident regions. According to the results we mentioned before, the level of social capital prevailing in the region where the bank branch granting the loan is based is expected to influence the loan officers' trusting attitudes as well as their evaluations of the risks of moral hazard that may occur with borrowers (affecting the price of credit risk, in turn). We estimate the effect that the social capital prevailing in the areas where the bank branches granting the loans are based plays on contracts stipulated with borrowers located in different regions. In these regressions it is possible to insert local-level fixed effects, associated with the region where the borrower is based, which absorb any other unobservable factor potentially relevant at the local level. In this way, and similarly to the strategy adopted by Guiso et al. (2004), we can address the critiques based on the omission of local variables. The second problem arises from the fact that expectations of firm performance, incorporated in financial outcomes, might influence family members' decision whether or not to keep control of the firm (Bennedsen et al., 2010; Miller et al., 2007). This point is addressed through Heckman's (1979) two-step regression framework, where family control is determined endogenously in the model.

1

Trust has been defined as an environmental characteristic which facilitates transactions in markets exposed to agency costs (Guiso et al., 2004).

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

3

The supply-side effects of the recent global financial crisis are investigated though a formal difference-in-difference approach à la Khwaja and Mian (2008). The heterogeneous degree of vulnerability, across Italian banks, to the shock in the interbank market is used to identify how banks have responded to the collapse of Lehman Brothers. In order to totally absorb any demand-side effects we take advantage of the fact that Italian firms normally have more than one lender and run regressions over first-differenced data, which include firm-fixed effects. Finally, our dataset includes a small number of loan contracts, which are secured by real collateral or personal guarantees (such secured loans accounted for around 5% of all contracts). We adopt the method proposed by Berger and Udell (1995) to control for the risk that our estimates suffer from simultaneity bias due to the joint determination of these components of contracts. The method is briefly described in Section 2.5. The rest of the paper is structured as follows: Section 2 presents the hypothesis tested, describes the empirical model and shows our results; Section 3 concludes. 2. Empirical analysis 2.1. Data sources We need information about firms' ownership structure and managerial organization, firm balance-sheet data, lender–borrower relationships and social capital indicators. Accordingly, our dataset comes from five main sources: (i) the Efige Survey, (ii) the Balance Sheet Register, (iii) the Central Credit Register, (iv) the Bank of Italy Loan Interest Rate Survey, and (v) the indicators of trust, recently produced by the Bank of Italy (de Blasio et al., 2014). In order to obtain the dataset: i) we use firm micro-data obtained from the Efige survey,2 conducted on a representative sample of manufacturing firms. The dataset contains detailed information on firms' characteristics, such as governance and management structure, workforce composition, investment and innovation activities, internationalization strategies and financial structure. As for family control, the following question was asked: “Is your firm directly or indirectly controlled by an individual or a family-owned entity?” The answer to this question is the basis for defining our main variable of interest, namely whether the firm is a family firm or not. ii) We match the above firm-level information with the Balance Sheet Register data, available with an annual frequency for the period from 2005 to 2011. iii) We recover financial data using Bank of Italy information collected, with quarterly frequency, in the Central Credit Register, which provides detailed data on the loans granted by Italian banks to each borrower whose total debt from a single bank exceeds 30,000 euros. These data are merged with the Bank of Italy Loan Interest Rate Survey, which contains information on the interest rates charged on each loan granted by 213 banks accounting for over 90% of total outstanding loans. This information enables us to construct unique bank–firm relationships. We focus on revocable overdraft facilities and on accounts receivables, due to the standardized characteristics underlying this kind of (short-term) lending. Specifically, banks can easily renegotiate revocable credit lines at very short intervals, fully capturing the repricing associated with changes in risk perceptions. iv) Finally we exploit social capital differences across Italian regions, by integrating the above observations with the indicators for trust, recently processed by the Bank of Italy.3 Data on social capital and trust are highly persistent, by their nature. In our dataset, they are available with a breakdown by region and are time-invariant. After a filtering procedure, we end up with 1877 family firms and 597 non-family firms; overall our sample consists of more than 107,000 financial contracts relative to 28 quarters, from 2005Q1 to 2011Q4; these contracts survived after cleaning outliers from raw data.4 2.2. Description of variables In Table 1 we report details of the definition of the variables used in the empirical analysis. The key variables of this study are RATE, FAM and TRUST: RATE (loan-level) is the interest rate, including fees and commissions, charged on short-term loans, i.e. overdraft facilities and accounts receivable granted by the bank to the firm.

2 The Efige dataset is a newly collected cross-country dataset obtained from a survey of 15,000 manufacturing firms with at least ten employees. The data were collected within the Efige project – “European Firms in a Global Economy: internal policies for external competitiveness” – supported by the Directorate General Research of the European Commission. The dataset contains detailed qualitative and quantitative information covering different areas for seven European countries (Italy, Austria, France, Germany, the United Kingdom, Spain and Hungary). The survey has a cross-section dimension; it has been carried out once, in 2010. 3 See de Blasio et al. (2014) and Albanese et al. (2013). 4 Raw data reported by banks to the Central Credit Register and by Chambers of Commerce to the Cerved archives, respectively, were cleared of ‘severe’ outliers. These outliers make up less than 0.00015% (1.5 per million) of a Gaussian population and have substantial effects on means, standard deviations and other statistics.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

4

Contract (bank/firmlevel)

VARIABLES

DESCRIPTION

N.

RATE

Bank interest rates, including commissions, applied on short-term loans (%) Dummy equals to 1 if loan is secured by a real guarantee Dummy equals to 1 if loan is secured by a personal guarantee Size of the loan Share of the firm's debt hold by the main lender

107,201

COLLATERAL GUARANTEES

Relationship (firm-level) Environm. (regionallevel)

Firm (firm-level)

TRANCE TOP-LENDER'S SHARE MULTIPLE EDU GDP JUDICIAL_INEFFIC NORTH SOUTH FAM

FIRM AGE SIZE RISK OWNERSHIP CONCENTR3 GROUP STRESS CEO_ABROAD CEO_AGE CENTRALIZATION Trust (regionallevel)

TRUST1(FIRM)

TRUST1(BANK)

TRUST2(FIRM)

TRUST2(BANK)

Number of firm's creditors (units) Share of population holding a university degree in the region Regional per-capita GDP Length of civil proceedings in the region (days) in 2010 Dummy for the firm's geographical localization Dummy for the firm's geographical localization Dummy equal to 1 if firm is directly or indirectly controlled by an individual or family-owned entity Age of the firm (years) Firm's total assets in tsd of € Firm's probability of default (Altman's Z-score); range from 1 to 9 Share of the firms' capital held by the first three shareholders Dummy that equals 1 if the firm belongs to a group Utilized loan as a share of granted loan Dummy that equals 1 if current CEO has worked abroad for at least 1 year Age of current CEO (9 age classes) Dummy equal to 1 if strategic decisions are centralized in CEO's hands ‘Civic awarness’ in the region where the firm resides (number of individuals getting information about politics at least once a week, as a share of entire total population based in the region) ‘Civic awarness’ in the region where the bank resides (number of individuals getting information about politics at least once a week, as a share of entire total population based in the region) ‘Associationism’ in the region where the firm resides (number of individuals supporting associations, as a share of the population based in the region) ‘Associationism’ in the region where the bank resides (number of individuals supporting associations, as a share of the population based in the region)

MEAN

SD

10.0

MIN

P5

3.7

1.2

P25 4.3

P50 7.3

P75 9.7

P95 12.6

MAX 16.6

20.1

107,201

0.02

0.12

0

1

107,201

0.03

0.16

0

1

107,201 107,201

165,620 0.21

158,124 0.40

1 0

107,201 107,201 107,201 107,201

7.8 14.2 29,048 874

4.6 1.8 5606 174

107,201 107,201 107,201

0.69 0.12 0.78

0.46 0.32

107,201 107,201 107,201

32 6136 5.40

21 5431 1.49

107,201

92

15

1.0 9.4 15,516 625

10,000

2.0 11.2 16,987 625

50,000

5.0 13.0 27,593 714

104,000

7.0 14.2 29,882 925

250,000

10.0 15.5 32,188 952

500,000

17.0 16.4 33,749 1214

920,000 1 43.0 19.6 67,238 1492 1 1 1

0

1 152 1

9 1048 3

18 2279 4

29 4296 5

40 8079 7

66 17,757 7

160 30,535 9

3

60

90

100

100

100

100

107,201 107,201 107,201

0.15 0.5 0.14

0.36 0.4 0.34

0 0.0 0

107,201 107,201

5 0.82

1 0.38

107,201

0.41

107,201

1 1.6 1

0.0

0.0

0.4

1.0

1.0

1 0

3

4

4

5

6

7 1

0.04

0.27

0.30

0.41

0.41

0.44

0.45

0.47

0.41

0.04

0.27

0.30

0.41

0.41

0.44

0.45

0.47

107,201

0.22

0.05

0.07

0.10

0.19

0.24

0.25

0.28

0.35

107,201

0.22

0.05

0.07

0.10

0.19

0.24

0.25

0.28

0.35

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

Table 1 Descriptive statistics (loan-level data).

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

5

Table 2 Trust in Italian regions. Civic awareness indicates the number of individuals getting information about politics at least once a week, as a share of entire total population based in the region. Associationism indicates the number of individuals supporting associations, as a share of the population based in the region. The values of Civic awareness and Associationism range between 0 and 1.

Campania Calabria Sicilia Basilicata Puglia Molise Abruzzo Sardegna Marche Umbria Veneto Emilia Lazio Toscana Piemonte Lombardia Valle d'aosta Liguria Friuli Trentino

TRUST1

TRUST2

Civic awareness

Associationism

0.347 0.304 0.301 0.267 0.282 0.334 0.318 0.421 0.372 0.360 0.440 0.448 0.408 0.415 0.418 0.413 0.352 0.468 0.457 0.457

0.073 0.131 0.069 0.154 0.102 0.138 0.140 0.220 0.190 0.187 0.231 0.251 0.161 0.279 0.195 0.244 0.224 0.188 0.259 0.352

FAM (firm-level) is a dummy which equals 1 if the firm is directly or indirectly controlled by an individual or family-owned entity; 0 if otherwise. TRUST (regional-level) is proxied by civic awareness [TRUST1], which indicates the number of individuals as a share of the population based in the region obtaining information about politics at least once a week. Table 2 shows how this measure of social capital varies within Italy: it is lower in the South, a little higher in the Center and higher still in the North, presenting some variation even within these macro-areas. In fact the percentage of individuals obtaining information about politics is equal, for example, to 27, 28 and 30 in the southern regions of Basilicata, Puglia and Sicily, respectively; by contrast, in northern regions such as Piedmont (42), Friuli Venezia Giulia (46) and Trentino Alto-Adige (46) the percentages are higher. A further proxy for trust is associationism [TRUST2], which indicates the number of individuals supporting associations, as a percentage of the population based in the region. We exploit information on the level of trust prevailing both in the region where the borrowing firm is located and in the region where the bank branch granting the loans is located: TRUST1(FIRM) and TRUST1(BANK) indicate the level of civic awareness in the regions where the firm and the bank branch granting the loan are located. The same holds for associationism. Other variables are: Environmental variables. GDP (regional-level) indicates the regional gross domestic product. These data are published quarterly by the Italian National Institute of Statistics (Istat). JUDICIAL_INEFFICIENCY (regional-level) is a proxy for the inefficiency of the local judicial system. It is measured as the average length (in days) of ordinary civil proceedings — including disputes over contracts, property law, tort and corporate law. The variable is time-invariant and refers to 2010. We also construct the variable EDU (regional-level), which measures, for each Italian region, the share of population holding a university degree. Istat provides these data annually. Firm characteristics. Our dataset contains firm-level information. The variable RISK proxies the firm's probability of default as defined by the Altman Z-Score index. This variable synthesizes information on several firms' financial statements and is adopted by Italian banks to evaluate the riskiness of loans. It takes values from 1 to 9. Firms with Z-Score values between 1 and 3 are deemed to be “low-risk” borrowers; those in the 4–6 and 7–9 ranges are “medium-risk” and “high-risk”, respectively. FIRM'S AGE measures the age of the firm. As proxies of SIZE we look at sales, number of employees and firms' total assets. Firms' ownership concentration is taken into account both through the share of capital held by the first shareholder (OWN_CONCENTR1) and the cumulative share of the first three shareholders (OWN_CONCENTR3). Three binary dummies indicate whether strategic decisions are centralized in the CEO's hands (CENTRALIZATION), the firm belongs to a group (GROUP), and the CEO has worked abroad for at least 1 year in the past (CEO_ABROAD). Further, CEO_AGE classifies firms according to the age of their CEOs. Relationship-lending characteristics. The dataset includes variables which are analyzed by the literature on relationship lending (Boot, 2000). This line of research emphasizes the importance of stable and large-scale relationships between banks and firms in limiting informational asymmetries in the loan market. TOP-LENDER'S SHARE (bank/firm-level) captures the strength of the relationship between the main lender and the borrower. This variable is measured by the amount of loans granted by the firm's main lender as a share of the total debt generated by the firm. MULTIPLE is the number of creditors from which each firm borrows. If it proxies for the degree of competition in the lending market, this variable may be negatively correlated with interest rates; conversely, a positive link may be a symptom of the limited quality of a company, which is unable to borrow additional money Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

6

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Table 3 Characteristics of family firms (firm-level data). Family firms

Contract

Relationship Environment

Firm

Trust

LOAN RATE COLLATERAL GUARANTEES TRANCHE TOP LENDER'S SHARE MULTIPLE EDUCATION GDP JUDICIAL_INEFFICIENCY NORTH SOUTH AGE SIZE (TOTAL ASSETS IN TSD OF €) RISK OWNERSHIP CONCENTRATION 1 GROUP STRESS CEO_ABROAD CEO_AGE CENTRALIZATION TRUST1(FIRM)-CIVIC AWARENESS TRUST1(BANK)-CIVIC AWARENESS TRUST2(FIRM)-ASSOCIATIONISM TRUST2(BANK)-ASSOCIATIONISM

Non-family firms

n.

Mean

sd

n.

Mean

sd

t-Stat

1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877 1877

9.9 0.02 0.03 174,505 0.30 5.74 13.98 28,801 878 0.686 0.139 31 5410 4.750 57.536 0.11 0.37 0.05 4.58 0.85 0.408 0.408 0.217 0.218

2.50 0.07 0.10 95,879 0.25 3.53 1.49 6114 185 0.464 0.346 21 5260 1.606 25.461 0.312 0.24 0.15 1.19 0.35 0.04 0.04 0.06 0.05

597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597 597

9.8 0.01 0.03 190,087 0.30 5.84 14.07 29,193 894 0.673 0.131 28 5605 4.757 55.107 0.28 0.34 0.08 4.50 0.72 0.412 0.412 0.221 0.222

2.5 0.05 0.10 115,813 0.26 4.13 1.57 7046 179 0.469 0.338 18 5478 1.690 28.445 0.446 0.24 0.19 1.08 0.45 0.04 0.04 0.06 0.05

1.07 2.70*** 0.63 −2.58** 0.34 −0.42 −1.52 −1.13 −1.70 0.34 0.57 2.82*** −0.77 −0.16 1.29 −10.75*** 2.12** −2.66*** 1.56 7.41*** −2.00** −2.09** −1.35 −1.45

from the major bank and hence is forced to turn to other banks. STRESS (bank/firm-level) proxies financial tensions in a lender– borrower relationship and is measured by the amount of loans utilized by the firm as a share of the total credit granted to that firm. Contract characteristics. The dataset includes other characteristics of the financial contracts such as TRANCHE, which indicates the size of the loan granted by the bank to the firm. COLLATERAL and GUARANTEES are binary dummies that equal 1 if the loan is secured by real collaterals or personal guarantees, respectively. 2.3. Some facts on family firms Table 3 provides a summary description of the characteristics of our sample of firms. As a proxy for size, in the table we highlight firms' total assets but we also scrutinize their sales and number of employees (not presented but available upon request). All these variables indicate that family firms are smaller than non-family firms, although not significantly. Moreover, the average age of family firms is 31 years, which is slightly older than that of their peers (28 years). The indicator showing the (ex-ante) probability of firms' defaulting (RISK) – ranging from 1 to 9 (high risk) – does not exhibit heterogeneities across family and non-family firms and the average is equal to 4.7. As for the geographical location of firms, around 70% are in the North of Italy, 17% in the Center and 13% in the South. Table 3 also shows that family firms are more likely to be located in low-trust environments than in high trust-ones. This evidence is significant, especially when the civic awareness indicator is considered. Moving to the characteristics of the loans granted to firms, we note that those to family firms are slightly but significantly smaller than those to non-family firms. This difference could reflect the fact that family firms are relatively smaller in terms of total assets. Conversely, we do not observe statistically significant differences with respect to the interest rates applied, while the loan contracts of family firms appear more likely to be secured by collateral. Finally, neither the figures indicating the share of loans held by firms' main lender nor the number of banks from which each firm borrows present patterns specific to the class of family firms. Moreover, strategic decisions are more likely to be centralized in the CEO's hands in the case of family-controlled firms; the data also show that family firms' CEOs are less likely to have gained experience by working abroad in the past. Ownership concentration appears to be relatively high and uniform among family and non-family firms. Table 3 shows that the share of capital held by the first shareholder is more than 55% for both these classes of companies. Both family and non-family firms have individuals as their largest shareholders in most cases. Table 3a shows that an individual is the main shareholder for 90% of family firms while industrial firms and holding companies occupy this position in only a small fraction of such firms (respectively 3.5 and 4.9%). These figures are slightly different for non-family firms: the proportion of firms having individuals, industrial firms and holding companies as their largest shareholder is now 62, 12.5 and 17.2%, respectively.5 Finally, data show that the importance of banks and other financial institutions among the largest shareholders of our sample firms is negligible.

5 However, when we also give consideration to the identity of the second and the third largest shareholders, the differences between family and non-family firms diminish: the proportion of non-family firms having individuals as their second or third largest shareholders is 86% (for family-controlled firms it is 95%).

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

7

Table 3a Firm ownership (firm-level data). 1st shareholder

Individual/group of individuals Industrial firm Holding firm Bank or insurance company Priv. equity and venture capital Public entity Other

2st shareholder

3rd shareholder

Family

Non-family

Family

Non-family

Family

Non-family

89.18 3.52 4.95 0.05 0.27 – 2.02 100.00

62.14 12.56 17.25 0.34 0.67 – 6.70 100.00

95.08 2.19 0.79 0.24 0.30 – 1.40 100.00

86.09 5.44 3.23 0.60 0.40 – 4.23 100.00

95.61 2.29 0.48 0.10 – – 1.53 100.00

86.22 5.57 2.64 0.59 0.88 – 3.81 100.00

Table 3b Family control and group affiliation (firm-level data). Firm belongs to a business group

Family Non-family Total

# (%) # (%) # (%)

No

Yes, domestic

Yes, foreign

Total

1673 89 425 71 2098 85

189 10 116 19 305 12

15 1 56 9 71 3

1877 100 597 100 2474 100

Table 3c Firm ownership of group affiliated-enterprises (firm-level data). 1st shareholder Family Individual/group of individuals Industrial firm Holding firm Bank or insurance company Priv. equity and vent. capital Public entity Other Total

43.1 17.2 34.8

2nd shareholder Non-family

Family

3rd shareholder Non-family

Family

Non-family

82.1 6.2 4.9 1.9 1.9

61.6 18.2 11.1 1.0 1.0

88.4 6.8 3.9

0.5

12.8 31.4 48.3 0.6 1.7

63.9 16.4 9.8 3.3 1.6

4.4 100

5.2 100

3.1 100

7.1 100

1.0 100

4.9 100

Table 3b shows that 15% of our sample firms are affiliated with business groups. The group is Italian for 12% of the sample and foreign for the other 3%. Among family-controlled firms the percentage of those belonging to a business group decreases slightly to 11% (10% are Italian) while the proportion of non-family firms affiliated with a group is larger and equal to 28% (19% are Italian).6 We look at the owners of the 15% of firms that are part of a group (Table 3c) to investigate the characteristics of the groups which are in the data set.7 The proportion of the firms having industrial firms and holding companies as their largest shareholder is now equal to 17 and 34% for the class of family firms, and 31 and 48 for the class of non-family firms. Symmetrically, the importance of the individuals among the largest shareholders decreases from 90 to 43% for family firms (and from 62 to 13% for non-family firms). The relevance of banks and insurance companies remains negligible.

2.4. Testable hypothesis The aim of this paper is to determine how creditors evaluate the agency costs of debt issued by family firms in different trust contexts. We expect creditors to price the personal ties of family firms as valuable assets in low-trust contexts, i.e. where the higher riskshifting attitudes of borrowers might be limited by family ties; symmetrically, the value of family firms' personal ties is expected to diminish in areas with a high social capital. From this argument we derive the following testable hypothesis:

6 The correlation matrix, not presented but available upon request, shows a negative association between the variables FAM and GROUP reflecting the figures highlighted in Table 3b. 7 The Efige survey unfortunately does not contain other information which would allow one to analyze the characteristics of these groups, as several cells of the questionnaire are ‘missing’: information about the name of the group, whether the firm heads the group and/or is controlled by other firms and whether its policies are influenced by the group.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

8

Depvar. RATE Contract

Relationship Environment

Firm

Trust

COLLATERAL GUARANTEES TRANCHE LENDERS SHARE MULTIPLE EDUCATION GDP NORD SUD FAM FIRM AGE SIZE RISK OWN_CONCENTR3 GROUP STRESS TRUST1(FIRM) FAM*TRUST1(FIRM) TRUST1(BANK) FAM*TRUST1(BANK) TRUST2(FIRM) FAM*TRUST2(FIRM) TRUST2(BANK) FAM*TRUST2(BANK) REGIONAL FIXED EFF. R2

1

1C

−0.2707*** −0.2679*** 0.0016 −0.0643*** 0.0000*** −0.6442*** 0.5754*** −0.1532*** −0.0026*** −0.3328*** 0.3092*** 0.0042*** −0.5668*** 0.5073***

−0.3655*** −0.2765*** −0.2706*** −0.2580*** 0.0023 −0.0635*** 0.0000*** −0.6417*** 0.5885*** −0.1512*** −0.0026*** −0.3311*** 0.3102*** 0.0042*** −0.5631*** 0.5131***

0.21

0.21

2

2C

−0.2702*** −0.2634*** 0.0023 −0.0234* 0.0000***

−0.3597*** −0.2638*** −0.2700*** −0.2537*** 0.0028 −0.0234* 0.0000***

−1.9232*** −0.0029*** −0.3353*** 0.3090*** 0.0045*** −0.5644*** 0.5080*** −8.9225*** 4.3058***

−1.9021*** −0.0029*** −0.3336*** 0.3099*** 0.0045*** −0.5612*** 0.5138*** −8.9018*** 4.2597***

0.21

0.21

3

3C

−0.2710*** −0.2677*** 0.0018 −0.0608*** 0.0000*** −0.6311*** 0.6003*** −2.0665*** −0.0028*** −0.3326*** 0.3074*** 0.0044*** −0.5582*** 0.5055*** −3.8872*** 4.6772***

−0.3578*** −0.2746*** −0.2708*** −0.2580*** 0.0024 −0.0602*** 0.0000*** −0.6322*** 0.6190*** −2.0503*** −0.0028*** −0.3309*** 0.3084*** 0.0044*** −0.5545*** 0.5112*** −3.7949*** 4.6425***

0.21

0.21

4

4C

−0.2684*** −0.2763*** 0.0062 −0.0103 0.0002***

−0.3998*** −0.2711*** −0.2682*** −0.2660*** 0.0068 −0.0116 0.0002***

−1.3376*** −0.0027*** −0.3661*** 0.2955*** 0.0045*** −0.4962*** 0.4965***

−1.3314*** −0.0027*** −0.3644*** 0.2967*** 0.0045*** −0.4928*** 0.5026***

−6.7197*** 2.8926***

−6.6440*** 2.8819***

YES 0.22

YES 0.22

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

Table 4 Ho before the crisis: OLS regressions of bank interest rates on family business and trust. Table presents OLS regressions for 56,101 observations at firm–bank level relative to 2474 firms, 178 banks and 15 quarters spanning from 2005Q1 to 2008Q3. Each regression includes fixed effects at level of bank, sector and time. The table presents eight alternatives of the model; for each alternative two regressions were run that differ only for the inclusion in the latter of two additional variables, capturing personal guarantees and real collateral, respectively. Columns (4–4C) and (6–6C) include also regional fixed effects. The dependent variable is RATE, i.e. the interest rate charged on short-term loans granted by the bank to the firm in the quarter; FAM is a dummy equals to one if the firm is directly or indirectly controlled by an individual or a family-owned entity. TRUST1(FIRM) is the level of ‘Civic awareness’ prevailing in the region where the borrowing firm resides; TRUST1(BANK) is the level of ‘Civic awareness’ prevailing in the region where the bank branch granting the loan resides; TRUST2 (FIRM) is the level of ‘Associationism’ in the region where the FIRM resides; TRUST2 (FIRM) is the degree of ‘Associationism’ in the region where the bank branch granting the loan resides; GDP is the regional gross domestic product; JUDIC_INEFF. is the average length (in days) of ordinary civil proceedings in 2010; EDU is the regional share of population holding a university degree; RISK is a synthetic measure of the firm's probability of default: firms are clustered into 9 qualitative risk classes; AGE measures the firm's age. SIZE measures the firm's total assets; OWN_CONCENTR3 is the firm's share of capital held by the first three shareholder; CENTRALIZATION is a dummy equal to 1 if the firm's strategic decisions are centralized in the Ceo's hands; GROUP is a dummy equal to 1 if the firm is part of a holding group; CEO_ABROAD is a dummy equal to 1 if the firm's Ceo has worked abroad for at least 1 year; CEO_AGE is the age of the firm's Ceo; NORTH and SOUTH are dummy variables referred to the firm's geographical localization; TOP-LENDER'S SHARE is the amount of loans granted by the main lender as a share to the total amount of debt issued by the firm; MULTIPLE is the number of creditors from which the firm borrows; STRESS is the amount of utilized loans as a share of the total loans; TRANCHE is the size of the loan granted by the bank to the firm; COLLATERAL and GUARANTEES are dummies equal to 1 if the loan is secured by real or personal collateral, respectively; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Significance adjusted for serial correlation and heteroskedasticity through the Huber– White Sandwich Estimator for variance.

Depvar. RATE Contract

Relationship Environment

Firm

Trust

COLLATERAL GUARANTEES TRANCHE LENDERS SHARE MULTIPLE EDUCATION GDP NORD SUD FAM FIRM AGE SIZE RISK OWN_CONCENTR3 GROUP STRESS TRUST1(FIRM) FAM*TRUST1(FIRM) TRUST1(BANK) FAM*TRUST1(BANK) TRUST2(FIRM) FAM*TRUST2(FIRM) TRUST2(BANK) FAM*TRUST2(BANK) REGIONAL FIXED EFF. R2

5

5C

6

6C

7

7C

8

8C

−0.2689*** −0.2651*** 0.0047 −0.0312** 0.0000***

−0.3398 −0.2493*** −0.2688*** −0.2559*** 0.0053 −0.0312** 0.0000***

−0.2687*** −0.2768*** 0.0074* −0.0038 0.0002***

−0.3972 −0.2627*** −0.2685*** −0.2666*** 0.0080* −0.0051 0.0002***

−0.2702*** −0.2646*** 0.0025 −0.0263** −0.0001***

−0.3542 −0.2676*** −0.2700*** −0.2549*** 0.0031 −0.0263** −0.0001***

−0.2698*** −0.2659*** 0.0051 −0.0265** 0.0000***

−0.3424*** −0.2546*** −0.2697*** −0.2565*** 0.0057 −0.0265** 0.0000***

−1.1225*** −0.0031*** −0.3449*** 0.3049*** 0.0048*** −0.5492*** 0.5157***

−1.1030*** −0.0031*** −0.3431*** 0.3058*** 0.0048*** −0.5465*** 0.5211***

−0.9587*** −0.0027*** −0.3680*** 0.2948*** 0.0047*** −0.4995*** 0.4965***

−0.9390*** −0.0027*** −0.3663*** 0.2960*** 0.0047*** −0.4964*** 0.5024***

−1.6185*** −0.0029*** −0.3349*** 0.3078*** 0.0046*** −0.5565*** 0.5089***

−1.6055*** −0.0029*** −0.3331*** 0.3086*** 0.0046*** −0.5532*** 0.5147***

−1.0838*** −0.0031*** −0.3429*** 0.3048*** 0.0049*** −0.5555*** 0.5150***

−1.0649*** −0.0030*** −0.3411*** 0.3058*** 0.0049*** −0.5525*** 0.5205***

−8.1384*** 3.5385***

−8.1194*** 3.5122***

−7.5007*** 4.1289***

−7.4313*** 4.0530***

−7.7794*** 4.3524***

0.21

−7.6947*** 4.2733***

0.21

−8.1094*** 3.6333*** YES 0.22

−8.0497*** 3.5525*** YES 0.22

0.21

0.21

0.21

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

Table 4 (continued)

0.21

9

10

Depvar. RATE Contract

Relationship Environment

Firm

Trust

COLLATERAL GUARANTEES TRANCHE LENDERS SHARE MULTIPLE EDUCATION GDP NORD SUD FAM FIRM AGE SIZE RISK OWN_CONCENTR3 GROUP STRESS TRUST1(FIRM) FAM*TRUST1(FIRM) TRUST1(BANK) FAM*TRUST1(BANK) TRUST2(FIRM) FAM*TRUST2(FIRM) TRUST2(BANK) FAM*TRUST2(BANK) REGIONAL FIXED EFF. R2

1

1C

−0.2523*** −0.2861*** 0.0122** −0.1019*** 0.0000*** −0.3858*** −0.0447 −0.2931*** −0.0041*** −0.2534*** 0.3132*** 0.0063*** −0.3276*** 0.3196***

−0.1718 0.0489 −0.2522*** −0.2842*** 0.0121** −0.1020*** 0.0000*** −0.3848*** −0.0438 −0.2928*** −0.0041*** −0.2540*** 0.3139*** 0.0063*** −0.3309*** 0.3214***

0.19

0.19

2

2C

−0.2514*** −0.2770*** 0.0131*** −0.0821*** 0.0000***

−0.1822 0.0511 −0.2513*** −0.2750*** 0.0130*** −0.0822*** 0.0000***

−2.5155*** −0.0045*** −0.2562*** 0.3140*** 0.0067*** −0.3194*** 0.3185*** −5.2110*** 5.3888***

−2.5162*** −0.0045*** −0.2568*** 0.3148*** 0.0067*** −0.3229*** 0.3204*** −5.2242*** 5.3915***

0.19

0.19

3

3C

−0.2512*** −0.2842*** 0.0128*** −0.1035*** 0.0000*** −0.5018*** 0.2407 −2.5239*** −0.0043*** −0.2550*** 0.3137*** 0.0064*** −0.3141*** 0.3181*** −1.9234 5.4286***

−0.1704 0.0443 −0.2512*** −0.2822*** 0.0127*** −0.1035*** 0.0000*** −0.4994*** 0.2387 −2.5242*** −0.0043*** −0.2555*** 0.3144*** 0.0064*** −0.3173*** 0.3199*** −1.9539 5.4302***

0.19

0.19

4

4C

−0.2467*** −0.3119*** 0.0119** 0.0254 −0.0004***

−0.1738 0.0405 −0.2467*** −0.3099*** 0.0118** 0.0249 −0.0004***

−1.1629*** −0.0043*** −0.2754*** 0.2996*** 0.0066*** −0.3026*** 0.3197***

−1.1644*** −0.0043*** −0.2758*** 0.3004*** 0.0066*** −0.3057*** 0.3215***

2.6773* 2.1136**

2.6761* 2.1180**

YES 0.20

YES 0.20

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

Table 5 Ho during the crisis: OLS regressions of bank interest rates on family business and trust. Table presents OLS regressions for 46,471 information at firm–bank level relative to 2474 firms, 178 banks and 12 quarters spanning from 2008Q4 to 2011Q4. Each regression includes fixed effects at level of bank, sector and time. The table presents eight alternatives of the model; for each alternative two regressions were run that differ only for the inclusion in the latter of two additional variables, capturing personal guarantees and real collateral, respectively. Columns (4–4C) and (6–6C) include also regional fixed effects. The dependent variable is RATE, i.e. the interest rate charged on short-term loans granted by the bank to the firm in the quarter; FAM is a dummy equals to one if the firm is directly or indirectly controlled by an individual or a family-owned entity. TRUST1(FIRM) is the level of ‘Civic awareness’ prevailing in the region where the borrowing firm resides; TRUST1(BANK) is the level of ‘Civic awareness’ prevailing in the region where the bank branch granting the loan resides; TRUST2 (FIRM) is the level of ‘Associationism’ in the region where the FIRM resides; TRUST2 (FIRM) is the degree of ‘Associationism’ in the region where the bank branch granting the loan resides; GDP is the regional gross domestic product; JUDIC_INEFF is the average length (in days) of ordinary civil proceedings in 2010; EDU is the regional share of population holding a university degree; RISK is a synthetic measure of the firm's probability of default (annual frequency): firms are clustered into 9 qualitative risk classes; AGE measures the firm's age. SIZE measures the firm's total assets; OWN_CONCENTR3 is the firm's share of capital held by the first three shareholder; CENTRALIZATION is a dummy equal to 1 if the firm's strategic decisions are centralized in the Ceo's hands; GROUP is a dummy equal to 1 if the firm is part of a holding group; CEO_ABROAD is a dummy equal to 1 if the firm's Ceo has worked abroad for at least 1 year; CEO_AGE is the age of the firm's Ceo; NORTH and SOUTH are dummy variables referred to the firm's geographical localization; TOP-LENDER'S SHARE is the amount of loans granted by the bank as a share to the total amount of debt issued by the firm; MULTIPLE is the number of creditors from which the firm borrows; STRESS is the amount of utilized loans as a share of the total loans; TRANCHE is the size of the loan granted by the bank to the firm; COLLATERAL and GUARANTEES are dummies equal to 1 if the loan is secured by real or personal collateral, respectively; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Significance adjusted for serial correlation and heteroskedasticity through the Huber–White Sandwich Estimator for variance.

Depvar. RATE Contract

Relationship Environment

Firm

Trust

COLLATERAL GUARANTEES TRANCHE LENDERS SHARE MULTIPLE EDUCATION GDP NORD SUD FAM FIRM AGE SIZE RISK OWN_CONCENTR3 GROUP STRESS TRUST1(FIRM) FAM*TRUST1(FIRM) TRUST1(BANK) FAM*TRUST1(BANK) TRUST2(FIRM) FAM*TRUST2(FIRM) TRUST2(BANK) FAM*TRUST2(BANK) REGIONAL FIXED EFF. R2

5

5C

6

6C

7

7C

8

8C

−0.2511*** −0.2791*** 0.0134*** −0.0888*** 0.0000***

−0.1796 0.0525 −0.2510*** −0.2772*** 0.0133*** −0.0889*** 0.0000***

−0.2466*** −0.3116*** 0.0116** 0.0272 −0.0004***

−0.1712 0.0438 −0.2466*** −0.3096*** 0.0115** 0.0267 −0.0004***

−0.2513*** −0.2797*** 0.0131*** −0.0872*** 0.0000***

−0.1828 0.0437 −0.2513*** −0.2776*** 0.0130*** −0.0873*** 0.0000***

−0.2513*** −0.2804*** 0.0130*** −0.0872*** 0.0000***

−0.1807 0.0477 −0.2512*** −0.2784*** 0.0129*** −0.0873*** 0.0000***

−1.0584*** −0.0045*** −0.2571*** 0.3113*** 0.0067*** −0.3214*** 0.3216***

−1.0585*** −0.0045*** −0.2578*** 0.3120*** 0.0067*** −0.3249*** 0.3235***

−0.4513*** −0.0043*** −0.2744*** 0.2986*** 0.0066*** −0.3003*** 0.3195***

−0.4513*** −0.0043*** −0.2748*** 0.2993*** 0.0066*** −0.3035*** 0.3213***

−1.7286*** −0.0044*** −0.2554*** 0.3131*** 0.0067*** −0.3237*** 0.3215***

−1.7296*** −0.0044*** −0.2559*** 0.3138*** 0.0067*** −0.3270*** 0.3235***

−0.6887*** −0.0044*** −0.2546*** 0.3125*** 0.0066*** −0.3297*** 0.3219***

−0.6887*** −0.0044*** −0.2552*** 0.3132*** 0.0066*** −0.3330*** 0.3238***

−2.3302** 3.4644***

−2.3410** 3.4678***

−0.8651 1.7237**

−0.8685 1.7251**

−2.8565*** 3.3904***

0.19

−2.8626*** 3.3928***

0.19

0.8255 0.7071 YES 0.20

0.8259 0.7086 YES 0.20

0.19

0.19

0.19

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

0.19

11

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

Table 5 (continued)

12

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Ho. Controlling for local characteristics, the economic cycle, firm and bank characteristics, types of financial contracts, type of lender– borrower relationship, a family firm is granted a loan interest-rate discount with respect to a non-family firm and the size of this discount decreases with the level of trust prevailing in the region where the firm is based. 2.5. The baseline specification We start by verifying our testable hypothesis Ho. We run regressions of alternatives based on the following specification: 0 0 0 RATEði; jÞ ¼ α FAMð jÞ þ β FAMð jÞTRUST þ δ TRUST þ Hð jÞ η þ KðiÞ γ þ Pði; jÞ ζ þ ε1

ð1Þ

where i and j refer to the i-th bank and the j-th firm and H(j), K(i), and P(i,j) are vectors of covariates defined at firm, bank and bank/ firm level, respectively. Our empirical model controls for a number of phenomena potentially biasing our results: time-level dummies control for economic cycle effects while time-varying firm-specific variables, such as the Altman Z-score indicator, are used to capture borrower risk. Each regression also includes bank-level fixed effects, to control for potential heterogeneities among lenders, − and economic sector-level and time-level dummies, to control for heterogeneity at the level of economic branch and to avoid bias due to cyclical fluctuations in economic activity. The empirical support for hypothesis Ho requires: α b 0 and β N 0. A negative value for α indicates that family firms benefit from a loan interest-rate discount with respect to their peers in low-trust regions (as TRUST decreases to 0). A positive value for β indicates that the discount depends on trust and would have been smaller if the same firms had been based in high-trust regions. This study analyzes the period from 2005Q1 to 2011Q4. The pre- and post-crisis periods are analyzed separately in this section in order to avoid structural breaks associated with the collapse of Lehman Brothers. A formal investigation of the supply side effects produced by the global financial crisis on the financing costs for family firms is carried-out in Section 2.9. As already noted, we need to control for simultaneity bias due to the presence in our dataset of a small number of loans secured by real collateral (COLLATERAL) or personal guarantees (PERSONAL). As for the regressions on loan interest rates, simultaneity bias might occur as COLLATERAL and PERSONAL might be seen as implicit components of the overall cost of financing and their presence in loan contracts is likely to be jointly determined with loan interest rates. To control for simultaneity bias we adopt the strategy followed by Berger and Udell (1995). For each alternative of our empirical model we run two regressions having the same set of covariates but with the variables COLLATERAL and PERSONAL included in the second specification only. The effects of the simultaneity bias would be negligible if the parameters of our key variables were highly comparable in the two specifications. 2.6. Econometric results: testing Ho before the crisis Table 4 presents eight alternatives of model 1 for the period prior to the Lehman crisis, from 2005Q1 to 2008Q3. As mentioned earlier, for each alternative we run two regressions that differ only for the inclusion in the latter of two variables, capturing personal guarantees and real collateral, respectively. We first discuss the key variable of this study and then present the main results for the remaining covariates. In column (1C), the estimated coefficient for FAM is significant and equal to −0.15. This means that, on average, the loan interest rate charged to a family firm is 15 basis points lower than the rate applied to a non-family firm, all other things being equal. The coefficient is almost the same when we exclude the implicit price components represented by guarantees and collateral (column (1)). In column (2C) we expand the model through the inclusion of the term FAM*TRUST(FIRM). The estimate for FAM is now −1.9 and the coefficient for FAM*TRUST(FIRM) is 4.3. These values highlight the relationship existing between the size of discounts to family firms and the level of trust of the region where the firm is located. More specifically, our estimates show that in a low-trust region, such as Basilicata, family firms benefit on average from a loan interest-rate discount of about 80 basis points compared with nonfamily firms; instead, in a high-trust region, such as Trentino-Alto Adige, this benefit drops to 5 basis points.8 Again, there is no difference in the size of the coefficient even if we leave the guarantee and collateral variables out of the regression (column (2)). As said, without controlling for trust components, in Italy the discount applied to family firms is equal, on average, to 15 basis points; this scale is comparable with the results obtained by Anderson et al. (2003) for the US, who find that family firms are charged a 32 basis point lower cost of debt financing relative to non-family firms (they also find that the discount rises up to 43 basis points for family firms where the ownership concentration is lower than 12%). The size of the discount (80 basis points) applied in Basilicata, the area having the lowest level of trust among Italian regions, should be evaluated by taking into account the relatively high dispersion exhibited by the distribution of interest rates in that region (the interquartile range equals 620 basis points). The higher levels of counterparty risk faced by lenders in low trust-regions (see also Section 2.9.5) might have enhanced the sensitivity of the banks' pricing behavior to family firms, in order to mitigate the agency conflicts existing in those areas. In column (3C), the binary dummies NORTH and SOUTH are included in the model. The sign and significance of the coefficients for FAM and FAM*TRUST(FIRM) remain confirmed. In columns (4) and (4C) we apply an empirical strategy à la Guiso et al. (2004) to address the risk that omitted components at a regional level are overestimating the effect of trust. As we said in the introduction, the inclusion of regional fixed effects, 8 Formally, the size of the discount applied to family firms is calculated as follows: Discount(trust) = −{E [loan rate | FAM = 1, trust] − E [loan rate |FAM = 0, trust]} = −(α + β trust).

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

13

associated with the region where the firm is located, permits to control for unobserved local components. To avoid the problem of perfect collinearity between the regional-specific trust variables and regional fixed effects, we focus on contracts signed between firms and lenders located in different areas and verify whether decisions on pricing are affected by interpersonal trust prevailing in the area where the bank branch granting the loans is based. FAM is still negatively associated with RATE and the coefficient for the interaction FAM*TRUST(BANK) is positive and statistically different from zero. These values indicate that the discount applied to a family firm decreases with the level of trust prevailing in the region where the bank branch granting the loan is located. The analysis is extended to a different proxy for trust in columns (5–5C) and (6–6C). Instead of civic awareness, we now look at the effect of the level of associationism prevailing, alternatively, in the region where the firm and the bank branch granting the loans are registered. The sign and significance of the coefficients are consistent with the results we obtained with the previous regressions (2–2C) and (4–4C), which consider civic awareness as the measure of trust. Lastly, in columns (7–7C) and (8–8C) we run the same regressions as in columns (4–4C) and (6–6C) but with the exclusion of the regional fixed effects. Again, the estimates for family firms and the interaction terms both retain their signs and significances. Moving to the analysis of the control variables, we find that contract characteristics also affect loan pricing. The estimate for TRANCHE is negative and statistically significant. Larger loans are associated with lower interest rates, as they are normally extended to firms having greater bargaining power with banks. The coefficient of TOP-LENDER'S SHARE is negative and significant. This means that loans are cheaper when the share of the firm's total indebtedness held by the firm's main lender is higher. This result may reflect the existence of either scale economies in monitoring activity or free-riding effects at work in loan markets characterized by ‘multiple lending’ structures (Carletti et al., 2007). The dummy NORTH is negative while the dummy SOUTH is positive; so that both are in line with the literature on Italian regional credit risk disparities. In particular, they show that a firm based in a northern Italian region pays 64 basis points less than a similar firm located in the Center; by contrast, a firm with its registered office in the South pays 59 basis points more. Moreover, our findings indicate a link between loan interest rates and education (computed at regional level), which is negative and statistically significant. We now consider our firm-level controls. All the coefficients are in line with our expectations. The estimate for RISK is positive and statistically significant. This means that riskier firms are charged higher rates by lenders. STRESS has a positive link with RATE: a firm having a high loan withdrawn-to-granted ratio pays relatively more than a borrower having a more balanced ratio. Further, loan pricing is negatively correlated with the age of the firm, its size and its belonging to a group. Finally, there is a positive relationship between the degree of ownership concentration and the cost of credit. 2.7. Econometric results: testing Ho during the crisis We enlarge our analysis to the years of the recent financial and sovereign crises (Table 5). The failure of Lehman Brothers exacerbated the information asymmetries affecting lending activity, while the sovereign crisis worsened funding conditions for banks, curbed economic activity and increased risk aversion among providers of financial services. For the purposes of this study, it is worth examining how the value of the network of personal ties exhibited by family firms, and priced by creditors, evolved in this very different context. In Table 5 we present the results of our regressions for the period from 2008Q3 to 2011Q4. As before we present eight alternatives of the model; for each alternative we run two separate regressions, which differ only for the presence in the second of two additional variables that measure the existence and value of personal and real guarantees, respectively. Column (1C) shows that family firms pay interest rates lower than those paid by non-family firms as the coefficient for FAM is significant and equal to −30. The discount is twice as large as that granted to family firms in the period preceding the failure of Lehman Brothers (see column (1C) in Table 4). Again, this coefficient does not change even when we remove the variables COLLATERAL and GUARANTEES from the model (column (1)). Column (2C) reports the estimates for the model augmented though the interactions between trust and family firms. The coefficient for FAM and FAM*TRUST(FIRM) are still significant and equal to −2.51 and 5.39. respectively, indicating that during the crisis the loan interest-rate discount applied to family firms continues to decrease with the trust prevailing in the region where the borrowing firm is located. The estimates presented in columns 3 through 8 show alternatives of the same model. In particular, we include the dummies NORTH and SOUTH in columns (3–3C), while we change our proxy for trust in columns (5–5C), (6–6C) and (8–8C). Moreover, in columns (4–4C) and (6–6C) we include regional level fixed effects à la Guiso et al. (2004) in order to test the role of the trust prevailing in the region where the bank is based. In all cases, our findings are qualitatively similar and provide support for the Ho hypothesis. Lastly, the significance and sign of all the coefficients we estimate for the control variables continue to be in line with our expectations. 2.8. Endogenous determination of family control This section addresses endogeneity concerns which might bias standard OLS estimates of the impact of family businesses on the cost of financing.9 The role of sample selection in generating spurious correlations between the indicator for family control and 9 We assume a random matching of lenders and borrowers, once we control for loan contract characteristics (size and collateral), heterogeneity across banks (through fixed effects) and across firms (the unobserved components of firm quality are also considered in the Heckman framework). Alternatively, the selection process of the partner – for both a lender and its client firm – could be specified by augmenting the Heckman framework through a two-sided matching model which would have to rely on adequate instruments. The set up of such a system is beyond the scope of this paper.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

14

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

firm outcomes has been recognized in the literature (Bennedsen et al., 2010). Endogeneity of family control with respect to economic or financial outcomes might arise as expectations regarding future earnings streams, equity market developments or financial burdens might affect the decisions of family owners, who might modify their equity stakes to the point of ‘abandoning the ship’ (Villalonga and Amit, 2006). To correct for endogeneity, we resort to the regression analysis of treatment effects (Heckman, 1979). In the first step of the empirical strategy we estimate a family-firm participation mechanism while in the second step we estimate the determinants of the financing costs — including a ‘selectivity correction’ among the covariates. The full model is specified as follows: 0

Prob½ FAMð jÞ ¼ 1jWð jÞ; TRUSTð jÞ ¼ θ TRUSTð jÞ þ W ð jÞp þ ε1

ð2Þ

0 0 0 RATEði; jÞ ¼ α FAMð jÞ þ βFAMTRUSTð jÞ þ δ TRUSTð jÞ þ H ð jÞη þ K ðiÞγ þ P ði; jÞζ þ ε2

ð3Þ

where i and j refer to the i-th bank and the j-th firm, respectively. W(j), H(j), K(i), and P(i,j) include covariates specific to the j-th firm, the i-th bank, and the relationship between the i-th bank and the j-th firm-levels, respectively. Each ‘treatment’ regression also includes fixed effects at bank, economic sector and time level. The participation equation includes economic sector level fixed effects. The exclusion restrictions, which are necessary for the identification of parameters, are met by including in vector W(j) three determinants of family control which do not enter the loan interest-rate equation. These covariates are the centralization of the strategic decisions in the CEO's hands [CENTRALIZATION], the level of the CEO's international experience [CEO_ABROAD] and the age of the CEO [CEO_AGE]. These variables should predict family control but not the interest rates charged on bank loans. Statistical evidence supports our assumption, since the correlation between RATE on the one hand and CENTRALIZATION, CEO_ABROAD and CEO_AGE on the other appears to be negligible at −0.008, 0.05 and 0.01, respectively. The inclusion of an interaction term FAM*TRUST in the loan interest-rate equation does not alter the exclusion restrictions needed for the identification of the parameters. It can be shown that models having interaction terms are identified when the same models without interactions are identified (Wooldridge, 2002 page 234). As in Section 2.5, for each alternative we run separate regressions (which are not presented here but are available upon request) that differ as regards the exclusion of the terms GUARANTEES and COLLATERAL. The estimates, which are not affected by the exclusion of these two variables, confirm that the effects of the simultaneity bias due to the joint determination of explicit and implicit components of the cost of financing are weak in our empirical model. 2.8.1. Results — treatment equation Table 6 presents the results of the Heckman treatment regressions of bank interest rates on family businesses and trust for the period 2005Q1 to 2008Q3. We run eight alternatives of our model. For each alternative, the upper part of the table reports the estimates for the equation that is our primary interest (RATE), while the outcomes of the first-stage probit regression are presented in the lower part of the table. In particular, regressions (4) and (6) correct for self-selection of statistical units in the category of family firms, and are also exempt from the risk of overestimating the effect of trust due to the omission of local factors as regional-level fixed effects are included (as in Guiso et al., 2004). To begin with, the estimate for LAMBDA is statistically different from zero in all the regressions, at the 1% probability level. This result confirms that estimates of single-equation models are subject to bias due to the omission of the component determining both family control and financial conditions. Turning to the equations that are our primary interest, the key variables FAM, FAM*TRUST1 (FIRM), FAM*TRUST2(FIRM), FAM*TRUST1(BANK) and FAM*TRUST2(BANK) are in line with the Ho hypothesis, as regards their sign and significance. In other words our findings indicate that in low-trust areas family firms are granted a loan interest-rate discount with respect to non-family firms; this discount decreases with the level of trust prevailing in the region where the firm is based or, alternatively, the region where the bank branch granting the loan is based. The comparison between the regression presented in Table 4 and the one presented in Table 6 shows that discounts applied to family firms are higher when family control is determined endogenously in the model. On the contrary, the Heckman and OLS frameworks do not return marked differences for the coefficients for the interaction terms FAM*TRUST(FIRM) and FAM*TRUST(BANK). Turning to our control variables, they all retain the explanatory power on bank interest rates that they showed in the previous standard OLS analysis. 2.8.2. Results — probit equation The lower part of Table 6 shows that family firms are more likely to be located in low-trust regions: in fact the coefficients for TRUST1(FIRM) and TRUST2(FIRM) are negative and statistically significant. As for environmental factors, we find that the presence of family firms is lower in regions characterized by higher educational levels and stronger economic growth. As regards firm-level characteristics, we expect family firms to be negatively associated with borrower risk. And in fact the coefficient for RISK is negative and statistically significant. The terms FIRM_AGE and CENTRALIZATION show that family firms are also slightly older and more likely to have adopted organizational structures where decision-taking is more centralized than in non-family firms. We also find a positive relationship between family control and ownership concentration, as indicated by the coefficient of Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

15

Table 6 Ho before the crisis: Heckman's treatment regressions of bank interest rates on family business and trust. Table presents Heckman's treatment regressions for 56,101 observations at firm–bank level relative to 2474 firms, 178 banks and 15 quarters spanning from 2005Q1 to 2008Q3. Each second-step (loan rate) regression includes fixed effects at level of bank, sector and time. Columns (4) and (6) include also regional fixed effects. Each first-step regression includes economic sector-level fixed effects. RATE is the interest rate on short-term loans granted by the bank to the firm in the quarter; FAM is a dummy equals to one if the firm is directly or indirectly controlled by an individual or a family-owned entity. TRUST1(FIRM) is the level of ‘Civic awareness’ prevailing in the region where the borrowing firm resides; TRUST1(BANK) is the level of ‘Civic awareness’ prevailing in the region where the bank branch granting the loan resides; TRUST2 (FIRM) is the level of ‘Associationism’ prevailing in the region where the FIRM resides; TRUST2 (FIRM) is the degree of ‘Associationism’ prevailing in the region where the bank branch granting the loan resides; GDP is the regional gross domestic product; JUDIC_INEFF is the average length (in days) of ordinary civil proceedings in 2010; EDU_LOCAL is the regional share of population holding a university degree; RISK is a synthetic measure of the firm's probability of default (annual frequency): firms are clustered into 9 qualitative risk classes; AGE measures the firm's age. SIZE measures the firm's total assets; OWN_CONCENTR3 is the firm's share of capital held by the first three shareholder; CENTRALIZATION is a dummy equal to 1 if the firm's strategic decisions are centralized in the Ceo's hands; GROUP is a dummy equal to 1 if the firm is part of a holding group; CEO_ABROAD is a dummy equal to 1 if the firm's Ceo has worked abroad for at least 1 year; CEO_AGE is the age of the firm's Ceo; NORTH and SOUTH are dummy variables referred to the firm's geographical localization; LENDER'S SHARE is the amount of loans granted by the bank as a share to the total amount of debt issued by the firm; MULTIPLE is the number of creditors from which the firm borrows; STRESS is the amount of utilized loans as a share of the total loans; TRANCHE is the size of the loan granted by the bank to the firm; COLLATERAL and GUARANTEES are dummies equal to 1 if the loan is secured by real or personal collateral, respectively; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Treatment regress. Depvar: RATE

1

2

3

4

5

6

7

8

Contract

−0.3460*** −0.2633*** −0.2704*** −0.2571*** 0.0028 −0.0725*** 0.0000*** −0.6714*** 0.6053*** −1.7235*** 0.0003 −0.3339*** 0.2978*** 0.0082*** −0.7973*** 0.5137***

−0.3411*** −0.2501*** −0.2699*** −0.2526*** 0.0034 −0.0299** −0.0001***

−0.3385*** −0.2605*** −0.2707*** −0.2568*** 0.0030 −0.0683*** 0.0000*** −0.6408*** 0.5985*** −3.6146*** 0.0001 −0.3337*** 0.2960*** 0.0084*** −0.7898*** 0.5118*** −4.1588*** 4.6091***

−0.3829*** −0.2590*** −0.2682*** −0.2650*** 0.0071 −0.0171 0.0002***

−0.3221*** −0.2364*** −0.2687*** −0.2548*** 0.0057 −0.0365*** 0.0000***

−0.3803*** −0.2505*** −0.2685*** −0.2656*** 0.0084* −0.0098 0.0002***

−0.3345*** −0.2537*** −0.2699*** −0.2539*** 0.0037 −0.0331*** −0.0001***

−0.3232*** −0.2408*** −0.2697*** −0.2554*** 0.0062 −0.0320** −0.0001***

−2.7125*** −0.0002 −0.3663*** 0.2863*** 0.0080*** −0.6941*** 0.5031***

−2.6186*** −0.0003 −0.3461*** 0.2942*** 0.0086*** −0.7697*** 0.5214***

−2.3320*** −0.0002 −0.3686*** 0.2855*** 0.0082*** −0.6988*** 0.5028***

−3.2384*** 0.0001 −0.3361*** 0.2961*** 0.0087*** −0.7901*** 0.5152***

−2.7072*** −0.0001 −0.3443*** 0.2933*** 0.0090*** −0.7906*** 0.5207***

COLLATERAL GUARANTEES TRANCHE Relationship TOP LENDER'S SHARE MULTIPLE Environm. EDUCATION GDP NORD SUD Firm FAM FIRM AGE SIZE RISK OWN_CONCENTR3 GROUP STRESS Trust TRUST1(FIRM) FAM*TRUST1(FIRM) TRUST1(BANK) FAM*TRUST1(BANK) TRUST2(FIRM) FAM*TRUST2(FIRM) TRUST2(BANK) FAM*TRUST2(BANK) REGIONAL FIXED EFFECTS First-step regression: probit(FAM) Trust TRUST1(FIRM) TRUST2(FIRM) Environm. EDUCATION GDP Firm RISK CEO_ABROAD CEO_AGE FIRM AGE OWN_CONCENTR3 GROUP CENTRALIZ. SIZE Lambda

−3.4551*** 0.0000 −0.3367*** 0.2978*** 0.0085*** −0.7940*** 0.5143*** −9.2311*** 4.2391***

−6.7953*** 2.9589***

−8.5284*** 3.6034*** −8.0466*** 4.3222*** −8.3347*** 3.6825*** YES

YES −0.9454*** −0.0147*** 0.0000*** −0.0259*** −0.0432** −0.0013 0.0074*** 0.0078*** −0.4322*** 0.3887*** 0.0052 0.9174***

−0.9454*** −0.0147*** 0.0000*** −0.0259*** −0.0432** −0.0013 0.0074*** 0.0078*** −0.4322*** 0.3887*** 0.0052 0.9110***

−0.9454*** −0.0147*** 0.0000*** −0.0259*** −0.0432** −0.0013 0.0074*** 0.0078*** −0.4322*** 0.3887*** 0.0052 0.9207***

−0.9454*** −0.0147*** 0.0000*** −0.0259*** −0.0432** −0.0013 0.0074*** 0.0078*** −0.4322*** 0.3887*** 0.0052 0.7873***

−7.8961*** 4.2146***

−0.9454*** −0.9593*** −0.0130*** 0.0000* −0.0260*** −0.0459*** −0.0006 0.0073*** 0.0077*** −0.4289*** 0.3869*** 0.0044 0.8778***

−0.9593*** −0.0130*** 0.0000* −0.0260*** −0.0459*** −0.0006 0.0073*** 0.0078*** −0.4289*** 0.3869*** 0.0044 0.7956***

−0.0147*** 0.0000*** −0.0259*** −0.0432** −0.0013 0.0074*** 0.0077*** −0.4322*** 0.3887*** 0.0052 0.9305***

−0.9593*** −0.0130*** 0.0000* −0.0260*** −0.0459*** −0.0006 0.0073*** 0.0078*** −0.4289*** 0.3869*** 0.0044 0.9369***

OWN_CONCENTR3, which is significantly different from zero. The parameter for CEO_ABROAD is negative and indicates that managers in family firms are less likely to have international experience than their peers in non-family firms. Finally, we find that firms belonging to a business group are less likely to be controlled by a family. The negative relationship between the variables GROUP and family control reflect the figures we discussed with reference to Table 3b. In Table 7 we repeat the regression analysis of the treatment effect for the period 2008Q4–2011Q4. The outcomes of this exercise are still in line with the Ho hypothesis tested in this study. 2.9. A formal analysis of the impact of the global financial crisis on the pricing of loans to family firms The results obtained in the previous section suggest that the size of the discount enjoyed by family firms increased during the recent global financial crisis. However, the supply-side effects associated with that event are not rigorously identified by our Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

16

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

regressions, which are run on separate time spans — i.e. for the periods before and after the failure of Lehman Brothers: other concomitant changes, such us variations in credit demand schedules, may have occurred after the shock and confound the banks' responses to the crisis. A difference-in-difference approach à la Khwaja and Mian (2008) is used to address this identification problem. More specifically, we exploit the heterogeneous vulnerability of Italian banks to the shock caused by the collapse of Lehman Brothers to compare, for the same firm, the changes in loan interest rates applied by banks exposed to a greater or lesser extent to that event. The inclusion of firmfixed effects – in first-differenced data – absorbs any time-varying and firm-specific components, such as potential variations in credit demand, thus providing a solution to our identification problem.

Table 7 Ho during the crisis: Heckman's treatment regressions of bank interest rates on family business and trust. Table presents Heckman's treatment regressions for 46,471 information at firm–bank level relative to 2474 firms, 178 banks and 12 quarters spanning from 2008Q4 to 2011Q4. Each second-step (loan rate) regression includes fixed effects at level of bank, sector and time. Columns (4) and (6) include also regional fixed effects. Each first-step regression includes economic sector-level fixed effects. RATE is the interest rate charged on short-term loans granted by the bank to the firm in the quarter; FAM is a dummy equals to one if the firm is directly or indirectly controlled by an individual or a family-owned entity. TRUST1(FIRM) is the level of ‘Civic awareness’ prevailing in the region where the borrowing firm resides; TRUST1(BANK) is the level of ‘Civic awareness’ prevailing in the region where the bank branch granting the loan resides; TRUST2 (FIRM) is the level of ‘Associationism’ in the region where the FIRM resides; TRUST2 (FIRM) is the degree of ‘Associationism’ in the region where the bank branch granting the loan resides; GDP is the regional gross domestic product; JUDIC_INEFF is the average length (in days) of ordinary civil proceedings in 2010; EDU is the regional share of population holding a university degree; RISK is a synthetic measure of the firm's probability of default (annual frequency): firms are clustered into 9 qualitative risk classes; AGE measures the firm's age. SIZE measures the firm's total assets; OWN_CONCENTR3 is the firm's share of capital held by the first three shareholder; CENTRALIZATION is a dummy equal to 1 if the firm's strategic decisions are centralized in the Ceo's hands; GROUP is a dummy equal to 1 if the firm is part of a holding group; CEO_ABROAD is a dummy equal to 1 if the firm's Ceo has worked abroad for at least 1 year; CEO_AGE is the age of the firm's Ceo; NORTH and SOUTH are dummy variables referred to the firm's geographical localization; TOP-LENDER'S SHARE is the amount of loans granted by the bank as a share to the total amount of debt issued by the firm; MULTIPLE is the number of creditors from which the firm borrows; STRESS is the amount of utilized loans as a share of the total loans; TRANCHE is the size of the loan granted by the bank to the firm; COLLATERAL and GUARANTEES are dummies equal to 1 if the loan is secured by real or personal collateral, respectively; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively.

Treatment regression. Depvar: RATE Contract COLLATERAL GUARANTEES TRANCHE Relationship TOP-LEND.'S SHARE MULTIPLE Environm. EDUCATION GDP NORD SUD Firm FAM FIRM AGE SIZE RISK OWN_CONCENTR3 GROUP STRESS Trust TRUST1(FIRM) FAM*TRUST1(FIRM) TRUST1(BANK) FAM*TRUST1(BANK) TRUST2(FIRM) FAM*TRUST2(FIRM) TRUST2(BANK) FAM*TRUST2(BANK) REGIONAL FIXED EFFECTS First-step regression: probit(FAM) Trust TRUST1(FIRM) TRUST2(FIRM) Environm. EDUCATION GDP Firm RISK CEO_ABROAD CEO_AGE FIRM AGE OWN_CONCENTR3 GROUP CENTRALIZ. SIZE Lambda

1

2

3

4

5

6

7

8

−0.0247 0.0959 −0.2648*** −0.2728*** 0.0089* −0.0967*** −0.0001*** −0.3481*** −0.0211 −3.4364*** 0.0017 −0.2147*** 0.2726*** 0.0141*** −0.6506*** 0.3226***

−0.0346 0.0998 −0.2642*** −0.2649*** 0.0098* −0.0798*** −0.0001***

−0.0225 0.0918 −0.2639*** −0.2707*** 0.0098* −0.0980*** −0.0001*** −0.4550*** 0.2502* −5.7024*** 0.0014 −0.2173*** 0.2734*** 0.0142*** −0.6324*** 0.3209*** −2.1694* 5.5967***

−0.0451 0.0919 −0.2585*** −0.2933*** 0.0084* 0.0106 −0.0004***

−0.0350 0.1021 −0.2647*** −0.2669*** 0.0103** −0.0827*** 0.0000***

−0.0419 0.0956 −0.2584*** −0.2930*** 0.008 0.0122 −0.0004***

−0.0367 0.0928 −0.2639*** −0.2671*** 0.0098* −0.0852*** −0.0001***

−0.0356 0.0981 −0.2647*** −0.2672*** 0.0100** −0.0806*** −0.0001***

−4.2572*** 0.0005 −0.2385*** 0.2663*** 0.0135*** −0.5845*** 0.3198***

−4.1843*** 0.0012 −0.2209*** 0.2710*** 0.0143*** −0.6341*** 0.3224***

−3.3809*** 0.0005 −0.2372*** 0.2642*** 0.0135*** −0.5844*** 0.3196***

−5.1218*** 0.0013 −0.2170*** 0.2731*** 0.0144*** −0.6384*** 0.3244***

−3.9872*** 0.0014 −0.2175*** 0.2715*** 0.0144*** −0.6476*** 0.3227***

−5.6751*** 0.0012 −0.2180*** 0.2739*** 0.0144*** −0.6361*** 0.3219*** −5.1962*** 5.5517***

−2.7946*** 4.1984***

2.3636 3.0434*** −4.0582*** 3.6611***

YES −0.8575*** −0.0133*** 0.0000*** −0.0437*** −0.0003 0.0237*** 0.0083*** 0.0077*** −0.3021*** 0.3935*** 0.0184*** 1.8233***

−0.8575*** −0.0133*** 0.0000*** −0.0437*** −0.0003 0.0237*** 0.0083*** 0.0077*** −0.3021*** 0.3935*** 0.0184*** 1.7993***

−0.8575*** −0.0133*** 0.0000*** −0.0437*** −0.0003 0.0237*** 0.0083*** 0.0077*** −0.3021*** 0.3935*** 0.0184*** 1.8077***

−0.8575*** −0.0133*** 0.0000*** −0.0437*** −0.0003 0.0237*** 0.0083*** 0.0077*** −0.3021*** 0.3935*** 0.0184*** 1.5730***

−2.3767*** 2.5531***

0.0764 1.6518** YES −0.8575*** −0.2651 −0.0146*** 0.0000*** −0.0446*** −0.0001 0.0242*** 0.0082*** 0.0077*** −0.3040*** 0.3903*** 0.0184*** 1.7838***

−0.2651 −0.0146*** 0.0000*** −0.0446*** −0.0001 0.0242*** 0.0082*** 0.0077*** −0.3040*** 0.3903*** 0.0184*** 1.5761***

−0.0133*** 0.0000*** −0.0437*** −0.0003 0.0237*** 0.0083*** 0.0077*** −0.3021*** 0.3935*** 0.0184*** 1.7975***

−0.2651 −0.0146*** 0.0000*** −0.0446*** −0.000 0.0242*** 0.0082*** 0.0077*** −0.3040*** 0.3903*** 0.0184*** 1.8101***

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

17

Table 8 Vulnerability of Italian banks: Net interbank position as percentage of total asset (percentages).Table shows percentiles of the distribution of the Net interbank position as a percentage of bank's total assets (VULN). VULN is equal to the sum of the differences between the outstanding amount of loans and the outstanding amount of deposits – both as a percentage of the bank's total assets – for the following four segments: Domestic Extra-, Domestic Intra-, Cross-Border Extra- and Cross-Border Intra-Groups. Positive (negative) values indicate that a bank is a net lender (borrower) in the interbank market. Percentiles

1%

5%

10%

25%

50%

75%

90%

95%

99%

2005 2008 2011

−12.4 −19.0 −34.5

−6.5 −14.5 −31.9

−5.1 −9.2 −19.4

−3.3 −3.8 −9.1

−0.4 −0.4 −1.6

0.2 0.4 0.2

2.4 2.9 1.7

5.3 4.9 4.6

12.2 10.9 6.2

As for the implementation of the Khwaja and Mian (2008) approach, we take advantage of the following three facts: – First, the shock produced by the 2007–09 financial crises was basically exogenous and took Italian lenders by surprise. This means that the observed impact of the shock should not be altered by preventive responses of Italian lenders to expectations of breaks. The crisis originated in the United States and affected Italian banks less than those of other countries thanks to their traditional intermediation model relying on stable funding from retail depositors and long-term relationships with borrowers. The Italian banking system was hit by the crisis primarily through the sudden collapse of activity in the interbank market, which caused a liquidity shock for banks' balance sheets; – Second, as in the case analyzed by Khwaja and Mian (2008), banks were exposed to the shock to a different extent in Italy, depending on their exposure in the interbank market. The banks relying to a greater extent on the wholesale market at the beginning of the crisis suffered more from the deterioration of funding conditions. In this respect, Bonaccorsi di Patti and Sette (2012) show that the contraction of credit after the collapse of Lehman Brothers was greater for banks that were more dependent on the wholesale markets, even after controlling for concomitant demand factors; – Third, in order to measure the differences in loan pricing across banks exposed to the shock to a different extent, we need to consider firms which are indebted (simultaneously) to more than one lender. This is the case of almost all firms in our sample. More generally, the typical indebtedness structure of Italian firms relies on the simultaneous presence of several financing banks, perhaps because this structure limits the power of the “inside” lenders to extract informational rents from their clients. Operatively, we consider all these facts and assess the bank's vulnerability to the shock by measuring how the bank was exposed in the interbank market at the onset of the crisis. The variable VULN measures the bank's net interbank position at the beginning of the crisis and considers all the segments of the interbank market. In particular, the net interbank position is obtained as sum of the differences between the outstanding amount of loans and the outstanding amount of deposits – both as a percentage of the bank's total assets – for the following four segments: Domestic Extra-, Domestic Intra-, Cross-Border Extra- and CrossBorder Intra-Groups. The segment Domestic Extra-Group includes the interbank transactions between domestic banks not belonging to any banking group or belonging to different banking groups; the segment Domestic Intra-Group includes transactions between banks belonging to the same group. The remaining two segments include the net positions existing vis-a-vis foreign entities.10 In 2008, the median of the VULN distribution was equal to 0%. As for banks in the 1st, 5th, 10th, and 25th percentiles, the values of VULN were negative and equal to −19%, −14.5%, −9.2% and −3.8%, respectively (Table 8). As for the event that allows us to divide the analysis into pre- and post-crisis periods, we take into account the common view looking at the failure of Lehman Brothers: this event generated panic within the financial community and exacerbated tensions in money markets. Therefore, for our baseline analyses we consider the period ending in September 2008 as the one preceding the crisis. In addition, we take into account the fact that liquidity premia started to fluctuate in European money markets in the second half of 2007, after the release of news on problem loans in the US mortgage markets. For this reason, we checked the robustness of our baseline analyses by shifting the end of the pre-crisis period back to June 2007. The following subsections present a non-parametric analysis of the supply-side effects of the 2007–09 liquidity shock on the pricing of loans to family firms and a formal exercise based on multivariate difference-in-difference regressions. 2.9.1. Bivariate analysis Table 9 presents bivariate mean difference-in-difference estimates for interest rates on bank loans charged to family and nonfamily firms, both before and after the financial shock, by high- and low-vulnerability banks. Banks whose net interbank position, as a percentage of total assets, was higher (lower) than the median of the distribution at the onset of the crisis, have been considered high- (low-) vulnerability banks. Moreover, we need to determine whether the supply-side effects of the shock vary with the level of interpersonal trust. Therefore, the test is carried out separately for high- and low- trust areas as well as for high- and low-vulnerability banks. In this exercise firms located in regions whose levels of trust are lower (higher) than the median value of the trust distribution were included in the low- (high-) trust class. 10 Liquidity in the interbank market dried up since August 2007, as banks preferred hoarding cash instead of lending it out even at short maturities (Heider et al., 2009). The liquidity drainage was primarily due to precautionary reasons in a context where lenders faced a sudden increase of credit risk (see also Acharya and Merrouche, 2013). Our indicator approximates the extent of these risks for each bank by aggregating the bank's net positions in all the segments of the interbank market. We also consider positions vis-a-vis foreign counterparts as cross-border linkages facilitated the transmission of the crisis to the financial markets around the world (see Milesi-Ferretti and Tille, 2011).

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

18

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Table 9 Spread on loan rates between family and non-family firms and vulnerable and non-vulnerable banks (percentages).Each cell reports the interest rate on loans applied to family firms minus the interest rate on loans applied to non-family firms. Banks whose net interbank position, as a percentage of total assets, (VULN) was – in June 2008 – higher (lower) than the median of the distribution are included among high- (low-) vulnerability banks. VULN is equal to the sum of the differences between the outstanding amount of loans and the outstanding amount of deposits – both as a percentage of the bank's total assets – for the following four segments: Domestic Extra-, Domestic Intra-, Cross-Border Extra- and Cross-Border Intra-Groups. Firms residing in regions whose levels of trust are lower (higher) than the median value of the trust distribution were included in the low (high) trust class. The pre-crisis and post-crisis periods include quarters from 2005Q1 to 2008Q3 and from 2009Q1 to 2011Q4, respectively. LOW TRUST

High vulnerability banks Low vulnerability banks Diff: high–low vulnerability

HIGH TRUST

Pre-crisis

Post-crisis

Diff: post-pre

Pre-crisis

Post-crisis

Diff: post-pre

−0.16 −0.38 0.22

−0.92 −0.24 −0.68

−0.76 0.14 −0.90

0.31 0.49 −0.18

−0.14 0.08 −0.22

−0.45 −0.41 −0.04

Table 9a Changes of banks' balance sheets between the pre-crisis (2005Q1–2008Q3) and the post-crisis (2009Q1–2011Q4) periods. This table shows percentiles of the distribution of banks' characteristics. PCT

5%

10%

25%

50%

75%

90%

95%

Δ (post–pre) capital/total assets Δ (post–pre) total assets (millions) Δ (post–pre) deposits/total assets

−0.02 −6898 −0.28

0.00 −2249 −0.16

0.00 −276 −0.03

0.02 33 0.03

0.03 305 0.11

0.05 1475 0.17

0.12 6340 0.23

Starting by considering low-trust contexts, we find that the discount on loan interest rates granted by high-vulnerability banks to family firms increased by up to 92 basis points after the shock, from 16 basis points in the pre-crisis period. By contrast, for lowvulnerability banks the discount granted to family firms decreased by up to 24 basis points after the shock, from 38 in the pre-crisis period. More importantly, the bivariate difference-in-difference estimate is equal to -90. These results indicate that in low-trust contexts the reaction registered by highly shock-exposed banks was significantly different from the response of less vulnerable banks; this suggests that the shock produced significant supply-side effects on the discounts applied to family firms. Conversely, when we look at firms located in high-trust regions, we obtain a different picture. Before the shock, all banks used to charge family firms higher rates; after the crisis, the difference between the loan interest rates applied to family firms and non-family firms decreased for both high vulnerability banks (−45 basis points) and low-vulnerability ones (−41 basis points). In a nutshell, these difference-in-difference estimates suggest two results: first, the supply-side effects of the shock on the pricing of loans to family firms seem to consist in an increase of the discount granted to family firms; secondly, this effect appears to be at work only in low-trust regions. 2.9.2. Multivariate analysis Bivariate analysis does not control for all the determinants of the changes in the discounts enjoyed by family firms after the shock, such as concomitant and heterogeneous variations of credit demand, which might confound banks' responses to the crisis. Following Khwaja and Mian (2008), we turn to multivariate difference-in-difference regressions in order to identify the supply-side effects. The change in loan interest rates before and after the collapse of Lehman Brothers, applied to the same firm by banks that were more or less exposed to the shock, is compared. To this end the time-series dimension of the original dataset is collapsed into just a pre-crisis and a post-crisis observation, defined from 2005Q1 to 2008Q3 and from 2009Q1 to 2011Q4, respectively. As a robustness check, we alternatively delimit the pre-crisis period to the third quarter of 2007, i.e. to when tensions in money markets materialized. The collapsed dataset is used to run alternatives of the following model11: Δpost−pre RATEði; jÞ ¼ FIRMð jÞ þ δ VULNðiÞ2008:q2 þ γVULNðiÞ2008:q2  FAMð jÞ þ Δpost−pre controlsðiÞ:

ð4Þ

To examine whether the supply-side response to the shock is different across family and non-family firms we interact the variable FAM(j) with our proxy for bank vulnerability measured at the beginning of the crisis.12 In this specification the coefficient for VULN indicates banks' response to the crisis event, i.e. to whether the pre-post changes of loan interest rates applied to a firm depend on the interbank exposure of the bank when the shock hit the interbank market; the parameter for the interaction VULN(i)*FAM(j) shows whether the response of a bank hit by the shock would have been different if the borrower had been a family firm. The inclusion of firm-fixed effects in first-differenced data regressions absorbs any variation in loan interest rates determined by changes in credit

11 Alternatively, the difference-in-difference estimator could be specified by including firm-quarter fixed effects; in this case we would get analogous parameters but less conservative standard errors. 12 The introduction, as a dependent variable, of the binary dummy FAM without interactions is not a feasible solution as, by construction, FAM is perfectly collinear with the firm-specific dummy variables and its effect would be totally absorbed by those dummies.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

19

Table 10 Difference-in-difference estimates: multivariate analysis — Pre-crisis period delimited to the second quarter of 2008. Data are collapsed in a pre-crisis and a post-crisis observation only, defined from 2005Q1 to 2008Q3 and from 2009Q1 to 2011Q4, respectively. Loan rate is the interest rate charged by bank i to firm j. VULN is the Net interbank position as a percentage of bank's total assets in June 2008. The variable is equal to the sum of the differences between the outstanding amount of loans and the outstanding amount of deposits – both as a percentage of the bank's total assets – for the following four segments: Domestic Extra-, Domestic Intra-, Cross-Border Extra- and Cross-Border Intra-Groups. Positive (negative) values indicate that a bank is a net lender (borrower) in the interbank market. BANK TOTAL ASSETS is the value of total assets of the lender; BANK CAPITAL is the equity capital ratio (equity capital as a share of total assets). TRANCHE is the size of the loan. Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Dep.var

LOW TRUST

HIGH TRUST

ALL

ΔPOST–PRE LOAN RATE VULN FAM*VULN ΔPOST–PRE BANK TOTASSETS ΔPOST–PRE BANK CAPITAL ΔPOST–PRE TRANCHE (ln) Const

OLS

FIRM FE

OLS

FIRM FE

OLS

FIRM FE

5.665*** −5.074** 0.000** 0.087 −0.284*** −1.274***

6.185** −6.472** 0.000 0.159 −0.223** −1.282***

−0.080 1.508 0.000 −0.245 −0.164*** −1.067***

1.311 −0.633 0.000 0.262 −0.161** −1.085***

2.567 −1.607 0.000* −0.241 −0.215*** −1.157***

3.962** −3.692* 0.000 0.318 −0.190*** −1.180***

Table 11 Difference-in-difference estimates: multivariate analysis — Pre-crisis period delimited to the second quarter of 2007. Data are collapsed in a pre-crisis and a post-crisis observation only, defined from 2005Q1 to 2008Q3 and from 2009Q1 to 2011Q4, respectively. Loan rate is the interest rate charged by bank i to firm j. VULN is the Net interbank position as a percentage of bank's total assets in June 2008. VULN is equal to the sum of the differences between the outstanding amount of loans and the outstanding amount of deposits – both as a percentage of the bank's total assets – for the following four segments: Domestic Extra-, Domestic Intra-, Cross-Border Extraand Cross-Border Intra-Groups. Positive (negative) values indicate that a bank is a net lender (borrower) in the interbank market. The data aggregate both national and foreign interbank positions. BANK TOTAL ASSETS is the value of total assets of the lender; BANK CAPITAL is the equity capital ratio (equity capital as a share of total assets). TRANCHE is the size of the loan. Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Dep.var

LOW TRUST

HIGH TRUST

ALL

ΔPOST–PRE LOAN RATE VULN FAM*VULN ΔBANK TOTASSETS ΔBANK CAPITAL Δln(TRANCHE) Const

OLS

FIRM FE

OLS

FIRM FE

OLS

FIRM FE

6.002*** −5.534*** 0.000* 1.532 −0.230** −1.165***

7.457*** −7.024** 0.000 1.669 −0.217** −1.163***

0.945 0.342 0.000 0.370 −0.207*** −0.869***

1.413 −0.580 0.000 −0.754 −0.229*** −0.866***

3.065* −2.313 0.000** 0.360 −0.217*** −0.995***

4.500** −3.705 0.000 0.654 −0.226*** −1.003***

demand schedules. Controls include indicators of bank characteristics potentially impacting on the changes in the interest rates applied by lenders (see Table 9a). As before, the test is carried-out separately for high- and low- interpersonal trust areas. We begin by presenting the results for firms located in low-trust regions (Table 10). The fixed-effect column indicates that the supply-side effects of the shock are significant. The coefficient for VULN is significantly positive and equal to 6.19 while that for VULN*FAM is significantly negative and equal to −6.47. These estimates indicate that a bank's one-standard-error increase in its net interbank position, as a percentage of total assets, translates into an increase in the loan interest rates applied to a non-family firm of 50 basis points and a decreases of 2 basis points when the client is a family firm.13 We get different results if we consider high-trust contexts. Now firm-fixed-effect regressions do not highlight any specific pattern for family firms for the pre–post change in loan interest rates. The coefficients for VULN and for the interaction VULN*FAM have the expected signs but they are no longer significant. Finally, when both low- and high-trust regions are examined jointly we consistently find that the estimates for our key variables lay in the middle between the coefficients we obtained before when we analyzed the two trust contexts separately.

13 The pre–post change of a one standard error increase in VULN is equal to (δ + γ)*SD.ERR(VULN) when the client is a family firm and δ*SD.ERR(VULN) when the client is a non-family firm. The SD.ERR of VULN is equal to 0.08 (8%).

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

20

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Fig. 1. Loan rates to family and non-family firms: low-trust areas. Figure presents interest rates (percentages) applied on loans. Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class.

As mentioned earlier, as a robustness check and in order to take into account the rise of liquidity premia in money markets as early as the second half of 2007, we repeated all these exercises considering a different dating of the pre- and post-crisis periods. Table 11 presents the results of regressions where the beginning of the liquidity shock is set at 2007Q3. The estimates are broadly similar to those we previously found-out. The change in the loan interest-rate discount granted to family firms is only significant in low-trust contexts: all other variables maintain their sign and significance. 2.9.3. Do pre-existing trends drive the changes in rates on loans granted to family firms after the shock? In principle, one might suppose that changes in the interest rates on loans granted to family and non-family firms after August 2007 simply reflect the evolution of different trends, which already existed before the shock. A non-parametric analysis is performed in order to verify the significance of this critique. Figs. 1 and 2 plot the evolution over time of the average loan interest rates applied to family and non-family firms in low- and high-trust regions. For the former, the figure shows similar pre-existing time trends for family and non-family firms; moreover, it can also be seen that loans granted to family firms are essentially cheaper. However, after the shock, and particularly from 2009 onwards, i.e. when the effects of the financial shock started to materialize in Italy, the two trends seem to be affected by different level shifts. This suggests that the abovementioned critique is not supported by empirical evidence. The same non-parametric analysis carried out in high-trust contexts shows similar pre-existing trends among family and non-family firms; it also shows that in this context loans granted to family firms are more expensive. Once again, we observe different level shifts: this means that in this case as well the critique is not supported by the data. 2.9.4. Controlling for group effects One might suppose that a family firm pays lower interest rates because both the firm and the lending bank belong to the same financial group. In this case the cheaper loans would not be associated with the family control's ability to reduce agency problems but might reflect group-level strategies compatible with the minimization of financing costs for a controlled entity. Moreover, a family firm could be part of a larger non-financial group having a larger quantity of (collateralizable) assets, which could be implicitly taken into account by lenders when evaluating the firm's creditworthiness. We verify whether our results are robust to these critiques by excluding all firms belonging to business groups from our original dataset and re-running the same set of difference-in-difference regressions. If the critiques were founded, the family-firm effects –

Fig. 2. Loan rates to family and non-family firms: high-trust areas. Figure presents interest rates (percentages) applied on loans. Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

21

Table 12 Controlling for group effects: Difference-in-difference estimates — Pre-crisis period delimited to the second quarter of 2008. Data exclude firms belonging to groups. Data are collapsed in a pre-crisis and a post-crisis observation only, defined from 2005Q1 to 2008Q3 and from 2009Q1 to 2011Q4, respectively. Loan rate is the interest rate charged by bank i to firm j. VULN is the Net interbank position as a percentage of bank's total assets in June 2008. VULN is equal to the sum of the differences between the outstanding amount of loans and the outstanding amount of deposits – both as a percentage of the bank's total assets – for the following four segments: Domestic Extra-, Domestic Intra-, Cross-Border Extra- and Cross-Border Intra-Groups. Positive (negative) values indicate that a bank is a net lender (borrower) in the interbank market. The data aggregate both national and foreign interbank positions. BANK TOTAL ASSETS is the value of total assets of the lender; BANK CAPITAL is the equity capital ratio (equity capital as a share of total assets. TRANCHE is the size of the loan. Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Dep.var

LOW TRUST

HIGH TRUST

ALL

ΔPOST–PRE LOAN RATE VULN FAM*VULN ΔBANK TOTASSETS ΔBANK CAPITAL Δln(TRANCHE) Const

OLS

FIRM FE

OLS

FIRM FE

OLS

FIRM FE

6.877*** −6.776*** 0.000* −0.162 −0.308*** −1.271***

8.423*** −9.862*** 0.000 −1.067 −0.279*** −1.241***

0.501 0.310 0.000 −0.797 −0.151** −1.103***

−0.183 −0.365 0.000 −1.114 −0.096 −1.119***

3.666** −3.221* 0.000 −0.532 −0.220*** −1.176***

4.869*** −5.719*** 0.000 −0.818 −0.180*** −1.178***

Table 13 Controlling for group effects: Difference-in-difference estimates — Pre-crisis period delimited to the second quarter of 2007.Data exclude firms belonging to groups. Data are collapsed in a pre-crisis and a post-crisis observation only, defined from 2005Q1 to 2008Q3 and from 2009Q1 to 2011Q4, respectively. Loan rate is the interest rate charged by bank i to firm j. VULN is the Net interbank position as a percentage of bank's total assets in June 2008. VULN is equal to the sum of the differences between the outstanding amount of loans and the outstanding amount of deposits – both as a percentage of the bank's total assets – for the following four segments: Domestic Extra-, Domestic Intra-, Cross-Border Extra- and Cross-Border Intra-Groups. Positive (negative) values indicate that a bank is a net lender (borrower) in the interbank market. The data aggregate both national and foreign interbank positions. BANK TOTAL ASSETS is the value of total assets of the lender; BANK CAPITAL is the equity capital ratio (equity capital as a share of total assets). TRANCHE is the size of the loan. Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Dep.var

LOW TRUST

HIGH TRUST

ALL

ΔPOST–PRE LOAN RATE VULN FAM*VULN ΔBANK TOTASSETS ΔBANK CAPITAL Δln(TRANCHE) Const

OLS

FIRM FE

OLS

FIRM FE

OLS

FIRM FE

5.912*** −5.271*** 0.000* 0.132 −0.306*** −1.280***

7.989*** −8.864*** 0.000 −0.845 −0.278*** −1.247***

0.843 −0.632 0.000 −0.838 −0.151** −1.109***

0.847 −1.693 0.000 −1.274 −0.096 −1.116***

3.285** −2.990* 0.000 −0.406 −0.220*** −1.182***

4.803*** −5.585*** 0.000 −0.729 −0.181*** −1.181***

i.e., the coefficient for the term FAM*VULN – would not be significant in low-trust contexts once firms belonging to groups were excluded from the dataset. Tables 12 and 13 report the results of this further robustness check. Again, as the starting date of the crisis, we take the failure of Lehman Brothers and subsequently the money market tensions observed since the third quarter of 2007. The results of these robustness checks are in line with our hypothesis. For firms located in low-trust contexts, the coefficients for the term FAM*VULN indicate a significant effect of family control in fixed-effect regressions. On the contrary, the parameters lose their significance once we look at borrowers located in high-trust regions. 2.9.5. Do the lower interest rates applied to family firms reflect corruption? To date, better conditions applied on loans extended to family firms have been supposed to depend on the power of family firms' personal ties to mitigate agency conflicts, especially in those environments where risks of managerial opportunism are remarkable due to penury of trust. However, there is also an alternative explanation, which states that family firms pay lower rates because lenders are corrupted or influenced in their credit risk assessment by the political power of family firms. On the connections between family firms and political power see Amore and Bennedsen (2013) and Xu et al. (2014). In other words, private benefits for bank directors, officers, or any individual having the power to influence the pricing of credit risk might induce these agents to overestimate the creditworthiness of a family firm and deliberately apply interest rates which are below market levels. In the case where corrupted lenders overestimate the ability of a family firm to repay loans, the (lower) interest rate should be in contrast with the actual riskiness of this borrower, as proxied by the ex-post observed loan performance. The significance of this alternative explanation is tested by exploiting information on problematic loans, which Italian banks have to report to the Credit Register.14 Table 14 presents some descriptive statistics on problematic loans observed for our sample firms 14 The analysis of problematic loans has been carried out over the universe of banks operating in Italy (around 650 banks in November 2013). On the other hand loan interest rates are reported to the Central Credit Register only by a sample of 213 banks, which cover around 90% of lending activity.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

22

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Table 14 Ex-post loan performance: bivariate analysis of problematic loans. Problematic loans include: i) bad loans (credits to an insolvent counterparty); ii) substandard loans (exposures to a counterparty facing temporary difficulties that are expected to be overcome within a reasonable period of time); iii) restructured loans (exposures in which the bank, as a result of the deterioration of the borrower's financial situation, agrees to change the original conditions, mainly by rescheduling deadlines or reducing interest rate) and iv) past due (exposures other than those classified above that are overdue by more than 90 days on a continuous basis). Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class. Family

Non-family

Problematic

Total

a (%)

Problematic

Total

b (%)

a−b −0.8*** −0.1 −1.5***

Low trust 2005–2007 2008–2010 2011–2013

183 244 555

7335 7584 6373

2.5 3.2 8.7

673 702 1861

20,637 21,380 18,170

3.3 3.3 10.2

High trust 2005–2007 2008–2010 2011–2013

458 746 1767

27,854 26,853 22,775

1.6 2.8 7.8

1421 2152 5268

87,011 85,379 74,721

1.6 2.5 7.1

0.0 0.3** 0.7**

from 2005Q1 to 2013Q3, the last period which was available in the Credit Register when the analysis was carried-out. The information is presented with breakdowns at the level of trust and type of firm, i.e. family or non-family. As explained in Section 2.9.1, firms located in regions whose levels of trust are lower (higher) than the median of the trust distribution are included in the low- (high-) trust areas. Table 14 highlights the ascending trends of the weight of problematic loans for both family and non-family firms, and in both high- and low-trust areas. In particular, during the period from 2005 to 2007 in low-trust contexts problematic loans represented 2.5 and 3.5% of all loans, for family and non-family firms, respectively. However, once the real effects of the global financial crisis hit the Italian economy, i.e. during the period from 2011 to 2013, the same figures reached the values of 8.7 and 10.2%, respectively. In high-trust contexts we observe similar rising values but in this case the differences between family firms and non-family firms are not remarkable. These figures – and particularly the different performances of loans granted to family and non-family firms in low-trust areas – do not provide evidence in favor of the corruption hypothesis: indeed they appear to be in line with our analysis of loan prices. We now pass to a multivariate analysis in order to formally examine the significance of the corruption hypothesis. We consider the loan-level binary dummy PROBLEMATIC LOAN (see footnote 12) and estimate the following probit model: 0

0

Prob½PROBLEMATIC LOANði; jÞ ¼ RISK ð jÞd þ W ð jÞp þ BANKðiÞ þ ε

ð5Þ

Table 15 Ex post loan performance: multivariate analysis of problematic loans Table presents probit regressions in which the dependent variable takes the value 1 if the loan is “problematic” (0 otherwise). Problematic loans are observed from 2005Q1 to 2013Q3. Risk class (nine risk classes) are binary dummy variables defined by the Altman Z-Score index. For the description of variables see Table 1. Problematic loans include: i) bad loans (credits to an insolvent counterparty); ii) substandard loans (exposures to a counterparty facing temporary difficulties that are expected to be overcome within a reasonable period of time); iii) restructured loans (exposures in which the bank, as a result of the deterioration of the borrower's financial situation, agrees to change the original conditions (mainly by rescheduling deadlines or reducing interest rate) and iv) past due (exposures other than those classified above that are overdue by more than 90 days on a continuous basis). Firms based in regions whose levels of trust are lower (higher) than the median value of the trust distribution are included in the low (high) trust class; asterisks denote significance at the 1%(***), 5%(**) or 10%(*) level, respectively. Low trust Dep var: problematic loans (0/1) Covariates Fam Risk class (Z-score) 2 3 4 5 6 7 8 9 AGE GROUP OWN_CONCENTR3 SIZE Bank FE Time FE Obs. Pseudo R2

High trust

All

−0.19***

−0.20***

−0.19***

0.00

−0.01

0.00

−0.04***

−0.05***

−0.03**

0.01 0.13 0.32*** 0.58*** 0.97*** 0.93*** 1.68*** 2.93***

0.01 0.07 0.27** 0.57*** 0.96*** 0.95*** 1.79*** 2.97***

−0.02 0.04 0.19*** 0.38*** 0.59*** 0.89*** 1.69*** 2.73***

−0.01 0.06 0.21*** 0.40*** 0.62*** 0.92*** 1.75*** 2.79***

−0.01 0.07 0.22*** 0.44*** 0.70*** 0.93*** 1.76*** 2.82***

Yes Yes 64,886 .27

Yes 280,127 .25

Yes Yes 270,140 .28

−0.03 0.07 0.13*** 0.42*** 0.57*** 0.87*** 1.75*** 2.72*** 0.00*** 0.06*** 0.00*** 0.08*** Yes Yes 222,207 .23

−0.02 0.07 0.22*** 0.43*** 0.67*** 0.90*** 1.69*** 2.76***

Yes 69,574 .21

0.10 0.25 0.50*** 0.84*** 1.18*** 1.17*** 2.13*** 3.00*** 0.00** −0.10** 0.00 0.13*** Yes Yes 53,145 .22

Yes 349,881 .23

Yes Yes 336,992 .27

−0.03 0.09* 0.19*** 0.48*** 0.68*** 0.90*** 1.81*** 2.75*** 0.00*** 0.01 0.00*** 0.09*** Yes Yes 277,067 .22

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

23

where i and j refer to the i-th bank and the j-th firm. RISK indicates the riskiness of the borrowing firm which is captured ex ante by the Altman Z-Score index. Nine risk dummies – specific for each value of the Altman Z-Score – are included in the model. The term W(j) includes other covariates specific to the j-th firm. Bank characteristics are controlled for by bank-level dummies. All the regressions include time-specific fixed effects. Table 15 presents the results. As expected, the likelihood of a problematic loan is monotonically (and positively) associated with the riskiness of the loans as indicated by the values of the dummies we estimate for each Altman Z-Score risk class. As for low-trust regions, the parameter for the dummy FAM is negative and significant in all the specifications. This means that a family firm has a lower probability of incurring repayment problems than a non-family firm. Moreover, loan performances appear to be lower for firms which have a more concentrated ownership and for those that are older. When we move to high-trust regions, we obtain different results. The coefficient for FAM is no longer significant. This means that where risks of managerial opportunism are lower, loan performances for family and non-family firms seem to be very similar. 3. Final remarks Family firms play a significant role in the productive system around the world (La Porta et al., 1999). In Italy – a paradigmatic case of heterogeneous distribution of social capital across regions – family firms account for almost 80% of businesses and their limited propensity to innovate is one of the drivers of the productivity gap exhibited by the Italian industrial system in international comparisons (Bugamelli et al., 2012). Our study sheds light on one of the ‘bright’ side of family businesses. It evaluates the agency costs of debt for family firms by comparing bank interest rates charged to family firms with those applied to their peers and measuring how this spread changes across contexts characterized by different levels of trust and after the eruption of the 2007–09 financial crisis. We find that family firms benefit from a loan interest-rate discount with respect to non-family firms. This discount: i) decreases considerably with the level of interpersonal trust prevailing where the contracts are stipulated and ii) is higher in low-trust regions after the collapse of Lehman Brothers. Our results support the thesis that the personal ties and incentives prevailing in family firms alleviates, in the perception of lenders, delegation problems and other conflicts affecting the governance of the firm and its creditworthiness. The contribution of family control to reducing the agency costs of debt is more valuable in areas where the lack of interpersonal trust enhances the tendency to behave opportunistically and the risk of expropriation for lenders. During bad times, the higher discount granted to family firms suggests that the family firms' long-term-commitments and reputation concerns might mitigate the temptation for companies to engage in risk-shifting high-expected returns strategies which might put at risk lenders' wealth, as a response to uncertainty or business difficulties (gambling for resurrection). Finally, we have no evidence of corruption effects on the part of lenders granting discounts to family firms. In fact we compare (ex-ante) loan riskiness and (ex-post) loan performance to show that the cheaper loans extended to family firms exhibit superior performances. Our findings are even stronger when we examine self-selection problems by modeling family control as an endogenous choice. Moreover, the role played by trust is not overestimated by the omission of local variables in our specification. We follow Guiso et al. (2004) and use regional-level fixed effects to absorb any other factor potentially responsible for the bias of our estimates. References Acharya, V., Merrouche, Ouarda, 2013. Precautionary hoarding of liquidity and interbank markets: evidence from the subprime crisis. Rev. Financ. (2012) 17, 107–160. Albanese, G., de Blasio, G., Sestito, P., 2013. Trust and preferences: evidence from survey data. Working papers, Number 911. Bank of Italy. Amore, M.D., Bennedsen, M., 2013. The value of local political connections in a low-corruption environment. J. Financ. Econ. 110 (2), 387–402. Anderson, R., Mansi, S.A., Reeb, D.M., 2003. Founding family ownership and the agency cost of debt. J. Financ. Econ. 68, 263–285. Banfield, E.C., 1958. The moral basis of a backward society. Free Press. Bennedsen, M., Pérez-González, F., Wolfenzon, D., 2010. The governance of family firms, corporate governance. In: Anderson, R., Baker, K. (Eds.), . Berger, N., Udell, G.F., 1995. Relationship lending and lines of credit in small firm finance. J. Bus. 68 (3), 351–381. Bonaccorsi di Patti, E., Sette, E., 2012. Bank balance sheets and the transmission of financial shocks to borrowers: evidence from the 2007–2008 crisis. Working Paper No. 848. Bank of Italy. Boot, A.W.A., 2000. Relationship banking: what do we know? J. Financ. Intermed. 9, 7–25. Bugamelli, M., Cannari, L., Lotti, F., Magri, S., 2012. The innovation gap of Italy's production system: roots and possible solutions, Bank of Italy, occasional papers. Carletti, E., Cerasi, V., Daltung, S., 2007. Multiple-bank lending: diversification and free-riding in monitoring. J. Financ. Intermed. 16, 425–451. Caselli, F., Gennaioli, N., 2013. Dynastic management. Econ. Inq. 51 (1), 971–996. D'Aurizio, L., Tommaso, O., Romano, L., 2015. Family firms, soft information and bank lending in a financial crisis (in this issue). de Blasio, G., Scalise, D., Sestito, P., 2014. Universalism vs. particularism: a roundtrip from sociology to economics. Occasional paper, no 212. Bank of Italy. Ellul, A., Guntay, L., Lel, U., 2007. External governance and debt agency costs of family firms. Board of Governors of the Federal Reserve System, International Finance Discussion Papers n. 908. Fukuyama, F., 1995. Trust: the social virtues and the creation of prosperity. Free Press. Guiso, L., Sapienza, P., Zingales, L., 2004. The role of social capital in financial development. Am. Econ. Rev. 94 (3), 526–556. Guiso, L., Sapienza, P., Zingales, L., 2008. Trusting the stock market. J. Financ. 63 (6), 2557–2600. Heckman, J., 1979. Sample selection bias as a specification error. Econometrica 47, 153–161. Heider, F., Hoerova, M., Holthausen, C., 2009. Liquidity hoarding and interbank market spreads. the role of counterparty risk. Working Paper Series No 1126. European Central Bank. Khwaja, A.I., Mian, A., 2008. Tracing the impact of bank liquidity shocks: evidence from an emerging market. Am. Econ. Rev. 98 (4), 1413–1442. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R., 1997. Trust in large organizations. Am. Econ. Rev. 87 (2), 333–338. La Porta, R., Lopez-de-Salinas, F., Shleifer, A., 1999. Corporate ownership around the world. J. Financ. 54 (2), 471–517. Milesi-Ferretti, Tille, 2011. The great retrenchment: international capital flows during the global financial crisis. Econ. Policy 26 (66), 289–346 (April 2011).

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006

24

M. Stacchini, P. Degasperi / Journal of Corporate Finance xxx (2015) xxx–xxx

Miller, D. Le, Breton-Miller, I., Lester, R.H., Cannella, A., 2007. Are family firms really superior performers? J.Corp. Financ. 13 (5), 829–858. Pinheiro, R., Yung, C., 2014. CEOs in family firms: does junior know what he's doing? Putnam, R., 1993. Making democracy work: civic tradition in modern Italy. Princeton University Press. Villalonga, B., Amit, R., 2006. How do family ownership, control and management affect firm value? J. Financ. Econ. 80 (2), 385–417 (May 2006). Wooldridge, J.M., 2002. Econometric analysis of cross section and panel data. The MIT Press, Cambridge, Massachusetts London, England. Xu, Nianhang, Yuan, Qingbo, Jiang, Xuanyu, Chan, Kam C., 2014. Founder's political connections, second generation involvement, and family firm performance: evidence from China.

Please cite this article as: Stacchini, M., Degasperi, P., Trust, family businesses and financial intermediation, J. Corp. Finance (2015), http://dx.doi.org/10.1016/j.jcorpfin.2015.01.006