Do distinct factors motivate financing decisions in family firms? Using debt to finance investment may raise a firm's bankruptcy risk, harming shareholders and consequently increasing the risk that the controlling family will be forced to relinquish control of the company. Consistent with this perspective, Crespí and Martín-Oliver (2015), Gallo, Tàpies and Cappuyns (2004), McConaughy Matthews, and Fialko (2001), and Agrawal and Nagarajan (1990) present evidence that family companies have less leverage than non-family companies, even when low leverage impairs their growth.

In contrast, Anderson and Reeb (2003b) and Anderson, Mansi and Reeb (2003) show no significant difference in the level of debt between family companies and non-family companies when the decision maker is concerned with business continuity. These researchers argue that although both types of companies follow a pecking order model of financing, they do so for different reasons: family companies issue equity only as a last resort because controlling families fear losing control of the firm (Blanco-Mazagatos, Quevedo-Puente, & Castrillo, 2007).

Finally, behavioral finance-based arguments may explain why family firms have higher levels of leverage. From this perspective, managers’ decisions are frequently biased by excessive optimism and overconfidence. These two biases are fundamentally different from each other, but generate the same result: the false perception that the firm is undervalued. While optimistic managers systematically overestimate their firm's cash flow (Heaton, 2002), overconfident managers underestimate earnings riskiness (Hackbarth, 2008), that is, the discount rates applied to firm's cash flows. Because of the perception that their firms are undervalued, optimistic and overconfident managers believe that external sources of financing, particularly new equity, are overpriced or “unduly costly” (Malmendier, Tate, & Yan, 2011, p. 1729), thereby making them to avoid equity issuances. If external finance is necessary, these managers will prefer to use debt first.

Although we do not directly test the effect of behavioral biases on firm leverage, optimism and overconfidence can possibly explain why the Brazilian firms in our sample are overleveraged. A previous study of Brazilian companies (Barros & Silveira, 2008) shows that most are classified as “biased” (p. 318) by excessive optimism and overconfidence is controlled by families. Based on this evidence and since our sample is very similar to that of Barros and Silveira, we assume that behavioral biases that possibly dominate our sample of family firms can at least partially explain our results. In light of this previous research, we propose:

Hypothesis 1. Leverage is higher in family firms than non-family firms.

The growing literature on capital structure adjustment speed estimates that companies modify their leverage to the target on a range that varies from 25% to 40% per year (e.g., Faulkender, Flannery, Hankins, & Smith, 2012; Flannery & Hankins, 2013; Flannery & Rangan, 2006; Huang & Ritter, 2009). Investigations in Brazil show similar results. Bogéa, Sheng, and Lora (2012), for instance, find speeds of adjustment as high as 24% for Brazilian firms, while Barros and Silveira (2008) find a range from 20% to 35%. While these studies help clarify the dynamics of firms’ financing decisions, the range is quite large, suggesting that some omitted factors are responsible for this high variance. Recent studies helped fill this gap by identifying some factors that influence firms’ capital structure adjustment speed, such as corporate governance features (Chang, Chou, & Huang, 2014), high bankruptcy costs (Elsas & Florysiak, 2011), evidence that companies with poor governance structures adjust faster when product market competition is more intense (Chang, Chen, Chou, & Huang, 2015), and a positive relationship with cash flow realization (Faulkender et al., 2012).

Ownership structure is another possible source of the differences in firms’ speed of adjustment. López-Gracia and Sánchez-Andújar (2007) show that leverage adjustment of small Spanish family firms is faster than of non-family companies, arguing that family companies’ lower leverage facilitates the raising of new debt and a swift adjustment to the capital structure target. They claim that the leverage of family companies is lower because family activism substitutes for debt as a governance mechanism to constrain the agency costs of equity, making debt unnecessary as a disciplinary device to control discretionary behavior. As agency theory suggests, debt is an efficient governance arrangement when free cash flow is high (Jensen, 1986).

However, our results regarding the level of debt contradict those of López-Gracia and Sánchez-Andújar (2007). We show that Brazilian family companies have more debt than non-family companies. Therefore, the governance substitution effect does not apply in the Brazilian case, thereby invalidating the argument of a faster speed of adjustment stemming from a low debt level. Consequently, we have to set an alternative explanation for the speed of adjustment of Brazilian family firms’ leverage.

When firms are over-leveraged, which is the case of family firms in Brazil, the capital structure is more likely to be adjusted through the issuance of new equity (Lyandres, 2010) instead of debt decreases. However, it is well known that the direct (i.e., gross spreads) and indirect costs (i.e., underpricing) of issuing equity are very high (e.g., Butler, Grullon, & Weston, 2005; Corwin, 2003; Lee & Masulis, 2009; Lyandres, 2010). These issuance (i.e., adjustment) costs make the capital structure balancing less frequent and, consequently, slower. In addition, raising external capital through debt would make family firms even more overleveraged. Overall, these facts suggest that family firms can be more financially constrained than non-family firms.

If family companies face more financial constraints, it likely costs them more to adjust their capital structure, and hence they take more time to do so. Korajczyk and Levy (2003) suggest that financially constrained companies make different decisions about capital structure compared to unconstrained firms. More specifically, financially unconstrained firms issue new equity when economic conditions are favorable (pro-cyclically, timing the market) and act counter-cyclically in issuing new debt. Thus, unconstrained firms can afford to deviate from their target leverage more often and revert faster and more easily.

In contrast, financially constrained firms cannot adjust their capital structure as fast as financially unconstrained firms can because it is more difficult for them to borrow, especially after adverse economic shocks (Korajczyk & Levy, 2003). They also have limited scope for issuing new equity. For this reason, constrained family firms tend to show a slower adjustment speed. Thus, we propose the following:

Hypothesis 2. Family firms have a slower capital structure adjustment speed than non-family companies.

Based on previous studies (e.g., Flannery & Rangan, 2006), we employ a partial adjustment model to estimate companies’ annual speed of adjustment to a target level of leverage, which, in turn, is a function of several lagged firm variables, represented by the vector X and its respective parameter θ in equation (1):

(1)

Flannery and Rangan (2006) argue that companies fail to adjust promptly to a target debt level due to high adjustment costs. Equation 2 represents the standard partial adjustment model that estimates firms’ annual speed of adjustment to the target leverage:

(2)

Substituting [1] into [2], we have:

(3)

Substituting (1 − ω) for β₁:

(4)

where γ is the parameter of the vector of variables that determines the target leverage and (1-β₁) is the estimated speed of adjustment to the target leverage. By definition, β₁ varies from 0 to 1.

To analyze the specific effect of family ownership on speed, we include an interaction variable between the binary variable fam, which takes a value of 1 for family firms and 0 otherwise, as well as lagged leverage. The parameter of this interaction is β₂, which indicates the partial effect of family ownership.

(5)

To account for unobserved heterogeneity and to address potential autocorrelation in the residuals, we use dynamic panel GMM estimators: Arellano and Bond's (1991) difference GMM and Blundell and Bond's (1998) system GMM.

We use data from publicly traded Brazilian companies for the period 2003-2013. We manually collected ownership and control data from mandatory reports that these firms are required to file with Securities and Exchange Commission of Brazil (Comissão de Valores Mobiliários [CVM]). We source the financial data from Economatica. In line with previous studies, we exclude firms belonging to banking and public utilities industries (e.g., Chang et al., 2014; Fama & French, 2002; Flannery & Rangan, 2006; Uysal, 2011) from the sample. The reason why banking companies, as financial intermediaries, are excluded from our sample relies on Fama and French (1992), who say that “the high leverage that is normal for these firms probably does not have the same meaning as for nonfinancial firms, where high leverage more likely indicates distress” (p. 429).

Public utilities, in turn, are excluded because their financial decisions depend on the regulatory environment within they work. For example, the Brazilian Electricity Regulatory Agency is the government agency responsible for supervising and authorizing companies to explore electricity generation, distribution, and transmission. Energy exploration licenses have defined terms (e.g., 30 years) that can be extended or not, thereby affecting firms’ investments and financing decisions, especially when the terms are about to expire. To decrease the effect of outliers, we winsorize all variables at the upper and lower one-percentiles. Our final sample consists of 1,937 firms-year observations for 257 firms from 19 industries.

Table 1 indicates that, regardless of the criterion used, Brazilian family firms tend to be older, smaller, and less profitable; they invest less, have fewer growth opportunities, have fewer tangible assets, have a lower depreciation index, and are more leveraged than non-family companies. The differences in their market leverage, in particular, are significant at the 1% level. Family companies’ higher leverage is also reflected in their book leverage (but only by criterion 1 and at the 10% level), which is measured as the ratio of total debt to total assets. We can also see that family firms predominate in our sample if we use criterion 1 (that allows a firm to be familiar even if the family has small shareholdings). Not surprisingly, the number of family firms decrease when we use criterion 2, which demands that a family has the control of the firm.

Table 1
Summary Statistics

Note. This table presents the summary statistics of family and non-family firms according to the two criteria presented in the section on measures. The sample is composed of all publicly traded companies in Bovespa (Brazil's stock exchange) in the 2003-2013 period, excluding firms from banking and public utilities industries. MktLev is the ratio of total debt to total debt plus market equity value. IND_LEV is the industry leverage ratio. BookLev is the ratio of total debt to total assets. EBIT_TA is the ratio of EBIT over total assets. LnTA is the natural logarithm of total assets. MTB is the ratio of the company's market value to the book value of the assets. DEP_TA is the ratio of depreciation allowances to total assets. FA_TA is the ratio of fixed assets to total assets. AGE is the age of the companies We winsorize all variables at the upper and lower one-percentiles. Criterion 1 distinguishes family firms from non-family firms by focusing on whether the LUS is a family regardless its shareholding, whereas criterion 2 on whether the LUS is a family who controls the company. Data are taken from Economatica, firms’ financial statements, and CVM reports. The column Difference reports the t-test for the difference of variables means and ***, ** and * represent statistical significance levels at one percent, five percent, and ten percent.

Table 2 shows the descriptive statistics of market and book leverages by industry (following Economatica classification), according to our criterion 1. To save space, we do not report the results using criterion 2, but they are similar. Except for banking and utilities, which are excluded from our analyses, one can see that our sample comprises firms from a wide range of industries. In addition, the distribution of family and non-family firms across these different industries seems to be proportional. Thus, it is unlikely that our main results stem from industry bias.

Table 2
Descriptive Statistics of Market and Book Leverages by Industry (Criterion 1)

Note. This table shows the descriptive statistics (mean and standard deviation) of market and book leverage by industry, according to Economatica classification. The sample is composed of all publicly traded companies in Bovespa (Brazil's stock exchange) in the 2003-2013 period, excluding firms from banking and public utilities industries. Market leverage (MktLev) is the ratio of total debt to total debt plus market equity value. Book leverage (BookLev) is the ratio of total debt to total assets. We winsorize all variables at the upper and lower one-percentiles. Criterion 1 distinguishes family firms from non-family firms by focusing on whether the LUS is a family regardless its shareholding, whereas criterion 2 on whether the LUS is a family who controls the company. Data are taken from Economatica, firms’ financial statements, and CVM reports. ***, ** and * represent statistical significance levels at one percent, five percent, and ten percent.

As a preliminary test, we analyze the extent to which family and non-family firms are far from their target leverage. First, we estimate the target leverage for each firm by regressing firms’ leverage against several firm characteristics and then taking the fitted values as the target leverage. This is a common strategy applied by previous studies (e.g., Dudley, 2012; Fama & French, 2002; Faulkender et al., 2012; Uysal, 2011) to estimate the unobservable target leverage. We particularly follow Dudley's (2012) procedure by running firm fixed-effect regressions in order to account for unobserved firm heterogeneity. In unreported analyses, we run several regressions and tests (e.g., Breusch-Pagan and Hausman tests), which indicated firm fixed-effects as the most appropriate estimator. We run two slightly different models with results in Table 3.

Table 3
Target Market Leverage (MktLev) Estimation

Note. This table presents the results from the estimation of the target leverage. The sample is composed of all publicly traded companies in Bovespa (Brazil's stock exchange) in the 2003-2013 period, excluding firms from banking and public utilities industries. Variables are defined in Table 1 and the section on measures. We winsorize all variables at the upper and lower one-percentiles. Data are taken from Economatica, firms’ financial statements, and CVM reports.***, ** and * represent statistical significance levels at one percent, five percent, and ten percent. The errors are clustered at firm level.

In Table 3, Model 1, we run a fixed effect (firm and year) regression with traditional determinants of capital structure commonly found in tests of speed of adjustment of leverage (e.g., Flannery & Rangan, 2006). Model 2, in turn, does not include the mean industry leverage as a regressor, but controls for firm fixed-effects and the interaction between industry and years. Regardless of the specification, the results are similar between the two models and the coefficients signs are consistent with previous studies. We avoid analyzing every single variable because our main interest relies only on the fitted values generated by these regressions. However, it is important to notice that the adjusted R² of both models are somewhat high, around 79%, suggesting that these models can explain a large portion of the leverage variance.

Table 4 shows the results related to the relationship between actual and target leverage of family and non-family firms, demonstrating that family companies are significantly overleveraged, whereas non-family companies are significantly underleveraged.

Table 4
t- test of the Difference Between Means of the Firm's Target and Actual Market Leverage

Note. This table presents the t-test of the difference between means of the company's target and actual market leverage. The sample is composed of all publicly traded companies in Bovespa (Brazil's stock exchange) in the 2003-2013 period, excluding firms from banking and public utilities industries. We winsorize all variables at the upper and lower one-percentiles. Criterion 1 distinguishes family firms from non-family firms by focusing on whether the LUS is a family regardless its shareholding, whereas criterion 2 on whether the LUS is a family who controls the company. Data are taken from Economatica, firms’ financial statements, and CVM reports.***, ** and * represent statistical significance levels at one percent, five percent, and ten percent.

Tables 1 and 4 provide initial findings that support our first hypothesis that family companies may have higher levels of leverage. However, one may argue that both the univariate statistics and the target leverage estimation ignore potential self-selection problems inherent to these analyses. All control variables we use to estimate the target leverage, for instance, may be endogenous to some extent. One way to address this issue is to test the effect of family ownership structure on the company's leverage using treatment effect techniques. Thus we run the “bias-corrected matching estimator” introduced by Abadie and Imbens (2011, p. 1) and Abadie, Drukker, Herr, and Imbens (2004).

Our interest result is the average treatment effect for the treated (ATET): the variation in leverage for the treatment group (family firms) contrasted to similar control companies (non-family firms). We allow for one matched observation per treated company and impose exact matching for industry classification. We use the Economatica classification scheme, resulting in 19 different industries (see Table 2 for all industries included in the analysis). As we run matching analyses with replacement, we allow control firms to be used for more than once, thus making it possible that one control firm matches two or more treated observations. In addition, we match firms by size, tangibility, profitability, market-to-book ratio, and age. Table 5 provides the results of this estimation.

Table 5
The Effect of Family Ownership on Leverage: Matching Estimation of Average Treatment Effects on the Treated (ATET)

Note. This table presents the matching estimator according to Abadie, A., Drukker, D., Herr, J. L., & Imbens, G. W. (2004). Implementing matching estimators for average treatment effects in Stata. The Stata Journal, 4(3), 290-311; and Abadie, A., & Imbens, G. W. (2011). Bias-corrected matching estimators for average treatment effects. Journal of Business & Economic Statistics, 29(1), 1-11. http://dx.doi.org/10.1198/jbes.2009.07333. We allow one matched observation per treated firms and impose exact matching for industry classification. Our matching variables are size, tangibility, profitability, market to book ratio, and age. The sample is composed of all publicly traded companies in Bovespa (Brazil's stock exchange), excluding firms from banking and public utilities industries. Data are taken from Economatica, firms’ financial statements, and CVM reports. We winsorize all variables at the upper and lower one-percentiles. Criterion 1 distinguishes family firms from non-family firms by focusing on whether the LUS is a family, and criterion 2 on whether the LUS is a family who controls the firm.***, ** and * represent statistical significance levels at one percent, five percent and ten percent.

As we can see from Table 5, family ownership is responsible for an increase in firms’ leverage, by 2.6 p.p. with the first criterion and by 3 p.p. using the second criterion. These results reinforce our first hypothesis and are quite different from most of previous studies. For example, Agrawal and Nagarajan (1990), McConaughy et al. (2001), Gallo et al. (2004), and Crespí and Martín-Oliver (2015) find that their sample of family companies have less leverage than non-family companies. The argument for family companies having lower debt levels is based on their aversion to financial distress and the resulting reluctance to borrow from banks (Gallo et al., 2004). It seems this rationale does not fit Brazilian family firms, since we find that their leverage is higher than for non-family firms. A possible explanation for the higher leverage of Brazilian family firms comes from behavioral economics. Barros and Silveira (2008) suggest that family firms tend to be biased by excessive optimism and overconfidence. These biases, in turn, can make the managers of family firms believe that their firms are undervalued and, consequently, that external finance are excessively costly. As a result, managers of family firms avoid issuing new equity and will prefer debt, if internal funds are insufficient to meet their investment needs.

Table 6 provides the results of the partial adjustment model of equation 5. The parameters in columns I and II refer to firms whose LUS is a family, and the coefficients in columns III and IV refer to firms whose LUS is both a family and the controlling shareholder. The coefficients in the odd columns were estimated by applying the difference GMM introduced by Arellano and Bond (1991) – henceforth AB – while the coefficients in the even columns were estimated by applying the system-GMM introduced by Blundell and Bond (1998) – henceforth BB. The results are consistent with our second hypothesis that family companies adjust to their target leverage significantly slower than non-family companies do.

Table 6
Speed of Adjustment to the Target Leverage

The parameters in column I suggest that non-family companies adjust their capital structure by 74.37% (1 – 0.2563) per year, whereas the value for family firms is 69.21% (1 – 0.2563 – 0.0516). Using the system GMM (BB) estimator, the difference is slightly greater: 48.52% (1 – 0.5148) for non-family firms and 40.25% (1 – 0.5148 – 0.0827) for family companies. All coefficients are significant at the 1% level. The difference in the AB and BB estimation results likely stem from the short sample period. According to Flannery and Hankins (2013), a short-panel bias may affect the consistency of the coefficients. After comparing certain estimators, they conclude that the BB estimator yields less biased coefficients. The Sargan test of overidentifying restrictions suggests that we cannot reject the hypothesis that the instruments are valid. Moreover, there is no second-order autocorrelation in the error terms, which validates the instruments used for both estimators (Baltagi, 2008).

Again, our results are quite different from that of previous studies. In the study by López-Gracia and Sánchez-Andújar (2007), the only one that compared the speed of adjustment between family and non-family firms, family firms show faster speed of adjustment toward target leverage, whereas we find slower speed for Brazilian family firms. López-Gracia and Sánchez-Andújar (2007) argue that the lower level of family firms’ debt can make it possible for these firms to use new debt to rapidly adjust their leverage. Since our sample of Brazilian family firms are over-leveraged, using new debt would make them to be even more leveraged. The adjustment, in this case, should be to decrease the leverage, perhaps with a new equity issuance. However, equity issuances are costly, making balancing of this nature less frequent and, thus, slowing the capital structure speed of adjustment.

The results remain qualitatively unchanged for the second criterion to define family firms. Columns III and IV show that these firms adjust leverage more slowly than non-family companies do. Employing the difference GMM estimator, the speed of adjustment is 73.32% (1 – 0.2281 – 0.0387) for family companies and 77.19% (1 – 0.2281) for non-family companies, while the system GMM estimator results are 40.63% (1 – 0.5579 – 0.0358) for family companies and 44.21% (1 – 0.5579) for non-family companies.

An alternative explanation that could be driving our results is related to the fact that, over time, some successful family firms have become non-family firms due to mergers, acquisitions, or new equity issuances. As we can see from Table 8, several family firms became non-family, and vice versa, within the period under analysis. For example, according to our criterion 1 (LUS is a family regardless its shareholding), 46 family companies became non-family companies and 68 non-family companies became family companies. Thus, one may argue that our analyses are merely contrasting capital structure issues between the most successful family firms of the recent past — that are no longer family firms — to those firms that remained as family firms and that are, possibly, less successful.

Table 8
Ownership Structure Shifts

Note. This table shows, year by year, the number of family firms that became non-family firms and vice versa. Criterion 1 distinguishes family firms from non-family firms by focusing on whether the LUS is a family regardless its shareholding, whereas criterion 2, on whether the LUS is a family who controls the company.

If the alternative explanation is true, neither the difference in leverage nor the difference in the speed of adjustment will be due to the family ownership, but to the firms’ efficiency instead. To rule out this possible explanation, we re-run our analyses by assuming that family firms do not shift to non-family structure, and vice versa, within the period of analysis. That is, a given firm is set as familiar (non-familiar) throughout the whole period if it first appears in our sample as a family company (non-familiar). Our main results hold. Table 9 presents the summary statistics and indicates that family firms are still more leveraged, older, less profitable, less tangible and smaller than non-family firms.

Table 9
Summary Statistics for Alternatives Criteria

Note. This table presents the summary statistics of family and non-family firms according to the two robust criteria presented in this section. The sample is composed of all publicly traded companies in Bovespa (Brazil's stock exchange) in the 2003-2013 period, excluding firms from banking and public utilities industries. Variables are defined in Table 1 and the section on measures. We winsorize all variables at the upper and lower one-percentiles. Rigorous criterion 1 distinguishes family firms from non-family firms by focusing on whether the LUS is a family regardless its shareholding in its first observation in our sample, whereas rigorous criterion 2 on whether the LUS is a family who controls the company when it first appears in our sample. Data are taken from Economatica, firms’ financial statements, and CVM reports. ***, ** and * represent statistical significance levels at one percent, five percent, and ten percent.

Regarding the distance from target leverage, Table 10 shows that, regardless of the criterion, family firms are still significantly over-leveraged while non-family firms are under-leveraged.

Table 10
t-test of the Difference Between Means of the Firm's Target and Actual Market Leverage

Note. This table presents the t-test of the difference between means of the company's target and actual market leverage. The sample is composed of all publicly traded companies in Bovespa (Brazil's stock exchange) in the 2003-2013 period, excluding firms from banking and public utilities industries. We winsorize all variables at the upper and lower one-percentiles. Rigorous criterion 1 distinguishes family firms from non-family firms by focusing on whether the LUS is a family regardless its shareholding in its first observation in our sample, whereas rigorous criterion 2 on whether the LUS is a family who controls the company when it first appears in our sample. Data are taken from Economatica, firms’ financial statements, and CVM reports.***, ** and * represent statistical significance levels at one percent, five percent, and ten percent.

Table 11 presents the estimation of the speed of adjustment. As in the prior subsections, the odd and even columns refer to the first-differences GMM and system GMM, respectively. As we can see, even using more rigorous criteria to define whether a firm is familiar or not, the results are supportive of our main ones.

Table 11
Speed of Adjustment to the Target Leverage Using Rigorous Family Criteria