Better safe than sorry: labor market regulation and cash holdings of founding family firms

“Revenue is vanity, profit is sanity, but cash is king” the saying goes. Indeed, firms rely on cash to finance their day-to-day operations and, thus, must maintain adequate cash reserves on their balance sheets. Arguably, firms decide the level of these cash holdings by trading off costs versus benefits (Bates et al., 2009; Harford et al., 2008; Pinkowitz and Williamson, 2007; Opler et al., 1999; Kim et al., 1998). Labor market regulation might affect this cost-benefit trade-off and has been identified as an important determinant of firms’ cash holdings (e.g., Beuselinck et al., 2021). However, while some authors document that firms increase their cash holdings when faced with more stringent labor market regulation (e.g., Karpuz et al., 2020), others find the opposite result (e.g., Klasa et al., 2009; Shin et al., 2018). In this paper, we argue that heterogeneity of shareholder preferences—in particular, socio-economic wealth considerations of family firms as pioneered by Gómez-Mejía et al. (2007)—may help partially explain these (presumably) contradictory findings. More specifically, we hypothesize that family firms increase their cash holdings relative to other firms in response to tightening labor regulations. Ultimately, we argue that the effect of labor market regulation on cash holdings identified in empirical studies depends on (among other aspects) the proportion of family firms in the sample.

Labor market regulation may impact firms’ cash holdings along two lines. First, tighter labor market regulation affects firms’ cost structures as labor costs become more sticky, and operating leverage increases as a result (Kahl et al., 2019; Mauer and Triantis, 1994; Mandelker and Rhee, 1984). Increasing operating leverage implies that firm earnings become more volatile, and firms are more likely to face financial constraints, and ultimately, higher bankruptcy risk (Claessens and Ueda, 2020). Thus, subject to tighter labor market restrictions, firms may increase their cash holdings to foster their financial flexibility (Karpuz et al., 2020) ^[1]. Second, firms might be reluctant to passively accept the increase in labor cost stickiness and react by establishing counter-measures. As such, they may reduce their cash holdings to improve their bargaining position (against employees), aiming to reduce the rent extraction potential of employees (e.g., Klasa et al., 2009; Shin et al., 2018). While the first perspective on firms’ response to labor market regulation is known in the literature as the “financial flexibility view”, the latter is often referred to as the “bargaining power view” (e.g., Chino, 2016; Simintzi et al., 2015; Matsa, 2010; Klasa et al., 2009).

Arguably, shareholder preferences might play a crucial role in the question of which of the two views will dominate in the firm-specific cost-benefit analysis, and thus, determine a firm’s response to labor market regulation. Indeed, Grossman and Stiglitz (1977) and others argue that shareholder unanimity (on a firm’s production and financing decisions) is only obtained under relatively strict assumptions. Yet, the existing literature has largely ignored firms’ ownership structures when examining the labor-finance linkage and assessing the explanatory power of both views ^[2]. We aim to partially close this gap by examining a specific type of shareholder: founding families.

The literature attributes very specific preferences to founding families, which are often referred to as socioemotional wealth (SEW) motives and the intention to preserve their legacy and ensure the business’s continued existence for the benefit of future generations (e.g., Berrone et al., 2012; Gomez-Mejia et al., 2011; Gómez-Mejía et al., 2007). This literature assumes that founding families have preferences for financial policies that safeguard firm continuity (e.g., Schmid, 2013), suggesting that founding family firms (FFFs) will increase their cash holdings in response to tightening labor market regulations. Moreover, it is argued that FFFs establish long-lasting and stable relationships with employees (Bennedsen et al., 2019; Huang et al., 2015; Lansberg, 1983, 1988; Sraer and Thesmar, 2007) as founding families view human capital as a relevant asset and not only as a counterpart claiming firms’ surpluses (see, e.g. Stavrou et al., 2007). This reasoning suggests that FFFs will react less to an increase in the labor force’s bargaining advantage and the associated rent extraction potential. Therefore, they are less likely to reduce the level of cash holdings in response to tightening labor market regulations. Finally, the literature he literature suggests that FFFs tend to exhibit risk-averse behavior (Anderson et al., 2012; Anderson and Reeb, 2004). As such, FFFs are expected to assign higher costs to financial risk compared to other firms. Consequently, they arguably have a strong interest in reducing financial risk and thus building up additional cash reserves in response to tightening labor market regulations. In summary, these arguments suggest that “financial flexibility” considerations might be more important for FFFs when deciding their response to labor market regulation, while “bargaining power” arguments might be more relevant for other firms. As such, we hypothesize that, relative to other firms, FFFs will, ceteris paribus, increase (decrease) their cash holdings in response to tightening (easing) labor market regulation.

To empirically test our assertion, we study a sample of 4,693 non-financial listed firms from 17 European countries over the 10-year period 2007–2016. We apply a four-step screening process to identify and classify founding family firms in our sample. First, we collect annual ultimate ownership information from Bureau van Dijk’s OSIRIS database using a threshold of 25% of the voting rights. Second, we inspect the time series of each firm and carefully cross-validate potential problems  in the initial OSIRIS data (e.g., when firm i has ultimate owner j in year $t−1$ and $t+1$⁠, but not in year t). Third, we identify all firms for which OSIRIS reports (at least once during the sample period) the presence of a family or individual as an ultimate owner. Fourth, we manually collect information for these firms on their founders, their relatives, and their (family) relation to the ultimate owner in the respective firm-year(s). Following Anderson and Reeb (2003) and others, we assign firms with an FFF status if there is a direct familial (consanguineous or marital) connection between the ultimate owner and the founder(s). In sum, this process provides us with a dummy variable indicating a firm’s (year-by-year) FFF status ^[3]. To proxy for frictions in the labor market, we use the data from the Economic Freedom of the World index provided by the Fraser Institute and construct a labor market regulation index (LMR) as an equally weighted average of the following sub-indicators: (i) the impact of minimum wages, (ii) hiring and firing practices, (iii) the share of the labor force whose wages are set by centralized collective bargaining, and (iv) the mandated cost of worker dismissal (see Section 2.2.2).

A simple sample-split analysis using mean comparisons shows that, in the full sample, cash holdings do not differ substantially between FFFs and non-FFFs. However, restricting the sample to high (low) LMR environments, we find that FFFs hold significantly more (marginally less) cash than non-FFFs. In high LMR settings, the magnitude of this difference is economically meaningful, amounting to approximately 14% of the average non-FFF’s cash holdings.

We confirm this pattern using regression analysis. Our baseline regression model explains the level of cash holdings while controlling for firm and industry-year fixed effects, a country-specific year trend, and a range of firm- and country-level characteristics. To account for potential correlations within countries, we cluster standard errors at the country level. To ensure the robustness of our results, we conduct several additional tests, detailed in the Online Appendix. First, we examine the sensitivity of our findings to different methods of estimating standard errors by re-estimating our model with alternative clustering approaches. Second, we employ an alternative regression specification that focuses on incremental cash holding decisions. Third, we control for potential confounding events and their interaction with the FFF variable to address concerns about omitted variable bias. Finally, we test the robustness of our results using alternative definitions of cash holdings and different measures of LMR.

Overall, we find compelling evidence that FFFs respond to tighter labor market regulation by increasing their cash levels relative to other firms, and some evidence that other firms reduce their cash holdings in response to stricter LMR. In terms of economic significance, a 1% increase in LMR materializes in a 0.083% increase in cash holdings in the average FFF, whereas the average non-FFF reacts with a 0.110% decrease in cash holdings (see Section 3.3). In sum, the average differential elasticity amounts to 0.193%, which results in a difference of 11.6% (7.7%) of the median (average) firm’s cash holding, assuming a shift from a 25%-LMR-quantile country to a 75%-LMR-quantile country.

To shed further light on the FFF-specific response to LMR and to strengthen our arguments, we also study the role of labor costs and the role of operating risk for FFFs’ response to LMR. The analysis of labor costs aims to strengthen the argument that the observed pattern is caused by variations in LMR. Consistent with the idea that country-specific LMR is particularly important for labor costs in labor-intense firms and domestic firms, we find that the FFF-specific response to LMR is more pronounced in these companies. The average differential elasticity increases from 0.193% for the average firm to 0.404% in labor-intensive firms and to 0.543% in domestic firms. Assuming a shift from a 25%-LMR-quantile country to a 75%-LMR-quantile country, the average differential elasticity corresponds to 22.7% (16.2%) and 36.6% (21.7%) of the median (average) cash holdings in labor-intensive and domestic firms, respectively. The analysis of operating risk reinforces the argument that risk mitigation and a precautionary demand for cash holdings drive the FFF-specific response to tighter LMR. Consistent with this, we find that the FFF-specific response to LMR is more pronounced in FFFs with business models characterized by higher operating risk.

While these findings already suggest a causal interpretation of the FFF-specific cash holdings response to LMR, we also directly address key identification challenges common to cross-country studies that may contest this interpretation (e.g., Bennedsen et al., 2019). First, we tackle the concern that FFF status is an endogenous choice by employing an instrumental variable (IV) approach. When instrumenting for FFF status, we find support for a causal interpretation of the FFF-specific response of cash holdings to LMR. Second, we guard against possible alternative explanations on firm-level for our findings. The main concern here is that systematic differences between FFFs and non-FFFs may drive the observed relationship in our baseline model. To address this, we conduct two robustness tests. First, we re-estimate our baseline model on a propensity score-matched sample of FFFs and comparable non-FFFs. Second, we run horserace regressions between FFF status and firm characteristics that plausibly correlate with financial flexibility considerations in the cash-labor relationship while also being linked to FFF status. In both cases, we continue to observe an FFF-specific response of cash holdings to LMR, reinforcing that our baseline results are not merely driven by alternative explanations.

Finally, we study the channels through which FFFs and non-FFFs adjust their cash holdings in response to tightening LMR. We find that non-FFFs respond to strengthened LMR by increasing dividends. In contrast, FFFs increase their cash flow from financing activities, in particular their debt issuance activities, while reducing dividends. Notably, neither FFFs nor non-FFFs seem to adjust their investment activities. These results further reinforce our main findings, suggesting that the observed LMR-induced changes in cash holdings stem from deliberate financial policy adjustments rather than being a mere byproduct of shifts in other corporate policies.

Our study contributes to various streams of literature. First, we add to the growing strand of research examining the relationship between labor market regulation and cash holdings. So far, the literature provides mixed evidence when examining bargaining power and financial flexibility arguments (e.g., Beuselinck et al., 2021; Karpuz et al., 2020; Cui et al., 2018; Shin et al., 2018; Ghaly et al., 2015; Schmalz, 2013; Klasa et al., 2009). In this study, we suggest that the cost-benefit analysis is very much firm-specific, and preferences of shareholders – specifically socioemotional wealth considerations of founding families – play a critical role in the question of which of the two views will ultimately dominate on a firm level. With empirical evidence supporting this argument, our study can help to reconcile previously ambiguous results, as our findings suggest that the overall effect identified in empirical research may depend on the proportion of family firms in the sample.

Second, we complement the growing literature on the behavior of family firms (Bennedsen et al., 2019; Chen et al., 2008; Wang, 2006). Our evidence corroborates socioemotional wealth arguments suggesting that family firms are more likely to focus on firm survival and retaining control over their business, rather than on mere wealth maximization which would require limiting the rent extraction potential of employees (see e.g. Breton-Miller et al., 2015; Berrone et al., 2012; Gomez-Mejia et al., 2011; Gómez-Mejía et al., 2007). Third, we contribute to the ongoing discussion about the independent role of cash holdings in corporate financial policy, providing supportive evidence for the theoretical predictions of Acharya et al. (2007) and Gamba and Triantis (2008) that firms facing operating or financial frictions prefer to save their internal funds to increase the cash buffer and reduce the risk of financial distress. Finally, our study complements research documenting that not only organizational form but also the type of blockholder matters when investigating corporate policy decisions (Glossner et al., 2020; Perotti, 1995).

The remainder of the paper is structured as follows: Section 2 introduces the data. Section 3 presents empirical results. Section 4 explores endogeneity concerns and alternative explanations. Section 5 examines the sources of cash increases. Section 6 concludes.

This section introduces the sample, the variable construction and presents descriptive statistics.

Our sample consists of all – active and inactive – publicly listed firms from 17 European countries ^[4], which are covered in OSIRIS and Datastream/Worldscope simultaneously for the period from 2007 to 2016. Following the standard approach, we drop observations with missing, negative or zero-values of total assets and total shareholders’ equity. We also exclude firms from the financial sector (Standard Industrial Classification (SIC) codes 6000–6999). Furthermore, we do not consider firm-years that have experienced reorganization ^[5]. Finally, we require non-missing contemporary values for the cash holding variables, the measure for labor regulation, as well as non-missing values for one-year lagged control variables.

The final sample is an unbalanced panel that includes (a maximum of) 28,113 firm-year observations from 4,693 unique firms. Founding family firms account for 6,215 (22%) firm-year observations, corresponding to 977 (21%) unique family firms in the sample.

Building on socioemotional wealth arguments, we opt for a relatively strict family firm definition that aims to ensure simultaneously:

• effective control of the family; and

• identification of the family with the company.

With regard to effective control, we apply a 25% threshold on voting rights as well as the condition of ultimate ownership (Bennedsen et al., 2019; Lins et al., 2010; Franks et al., 2012). To capture identification of the family with the company, we focus on founding families, i.e. founders and their relatives. Evidence for a strong relation of the founder to the firm can frequently be observed e.g. by the identity of the family name and the name of the company (Berrone et al., 2012; Gómez-Mejía et al., 2007; Westhead et al., 2001). In sum, we are interested in identifying firms, where the founding family is the ultimate owner of the firm at a threshold of 25%.

We apply a four-step screening process to identify such firms in our sample. First, we collect annual ownership information from Bureau van Dijk’s, a data provider well-known in the family firm literature (e.g., Bennedsen et al., 2019; Gnoth et al., 2024; Lins et al., 2013). To ensure that we are able to track complex ownership structures of European firms even throughout a network of multiple intermediate firms, we rely on the “Global ultimate owner” (GUO) variable ^[6] from the OSIRIS database using a threshold of 25%. Second, we inspect the time-series of each firm and carefully cross-validate anomalies in the initial OSIRIS data, i.e. when a firm i has an ultimate owner j in year $t−1$ and $t+1$⁠, but not in t^[7]. Third, we identify all firms with a person or family as global ultimate owner in any given year. Fourth, we manually collect founder information. To identify the founders of a company we use publicly available sources like annual reports, the company’s website or news articles. We then complement this by collecting information on the relation between the reported ultimate-owner and the company’s founders.

Ultimately, we classify a firm as a founding family firm in year t, if:

• the firm has an ultimate-owner possessing at least 25% of a firm’s voting rights in year t; and

• the ultimate-owner is related by blood or marriage to at least one of the founders of the company.

Based on this classification, we construct a dummy variable that is equal to one for founding family firms, and zero otherwise. In contrast to other studies building upon (partly) time constant family firm definitions (i.e. Schmid (2013); Pindado et al. (2011, 2012)) our approach provides us with a yearly varying FFF variable allowing us to use firm fixed effects regression models in the analysis to control for firm-specific heterogeneity.

To operationalize the strictness of labor market regulation in a particular country in our sample, we use sub-indicators on labor market regulations from the Economic Freedom of the World (EFW) index provided by the Fraser Institute ^[8] and widely used in labor economics studies (e.g., Edmans et al., 2020; Sturm and De Haan, 2015; Bernal-Verdugo et al., 2012; Feldmann, 2009; Siegel and Larson, 2009; Freeman et al., 2008) ^[9]. Since the EFW index is available for a long time period over 2000–2017 for most countries all over the world, it allows us to estimate the family-firm related relationship between cash holdings and labor for our entire sample. While labor market regulation is a multidimensional concept, the sub-indicators of the EFW^[10] cover various features and manifestations that are responsible for the frictions at the labor market. In our study, we focus on four sub-indicators referring to:

1. the impact of minimum wages;

2. hiring and firing practices;

3. the share of labor force, whose wages are set by centralized collective bargaining; and

4. the mandated cost of worker dismissal.

These four sub-indicators cover the dimensions that have been used by previous studies, examining the relation between firms’ financial policy and labor regulation, to proxy for labor market frictions ^[11].

Each sub-indicator is continuous and can take any values in the range from 0 to 10, with higher scores representing higher economic freedom and thus, less strict labor regulation. To benefit from higher variation, we aggregate four sub-indices to one index on the labor market regulation index (LMR) as an equally weighted average ^[12]. For ease of interpretation, we subtract the value of the index from 10, such that high values of the index now correspond to more rigid labor market regulations. As each sub-indicator varies within as well as across countries, our LMR index allows for time-series and cross-sectional analysis.

We define different measures of cash holdings. First, we follow Barbalau et al. (2022); Karpuz et al. (2020); Bates et al. (2009) and others and define Cash as the sum of cash and short-term investments divided by the book value of total assets. Second, we follow Opler et al. (1999) and define CashToNetAssets as the sum of cash and short-term investments divided by the book value of net assets, i.e. total assets net of cash and short-term investments. Since previous literature indicates that the latter measure might be prone to outliers (e.g., Bates et al., 2009; Foley et al., 2007), we use Cash as the primary and CashToNetAssets as an alternative measure. Finally, in unreported additional tests, we also measure cash holdings as the cash-to-sales ratio. Restricting the sample to firms with positive sales exceeding €1mio (e.g., Hovakimian and Li, 2011), the results of this exercise support the results presented below.

We add firm-level and country-level controls to our regression models. Firm-level controls are motivated by the studies of Opler et al. (1999); Song and Lee (2012); Graham and Leary (2018) and include: firm size (Size) defined as the natural logarithm of the book value of total assets; market-to-book-ratio (Growth) defined as the book value of total assets minus the book value of equity plus the market value of equity, divided by the book value of total assets; total debt to total assets (Leverage); net working capital to total assets (NWC); investments defined as the sum of capital expenditures, acquisitions, and R&D-expenditures to total assets (Investment); dividends defined as a dummy variable that equals one if a firm pays cash dividends in a particular year, and zero otherwise (⁠$DividendDummy$⁠); cash flow to total assets, where cash flow is defined as earnings after interest, dividends, and taxes but before depreciation (⁠$CashFlow$⁠); and industry cash flow volatility defined as the average of the firm’s standard deviation of cash flows for each country-industry (classified by 48 Fama-French industries) in a particular year, where the firm–year standard deviation of cash flow to assets is calculated for the previous five years (⁠$CashFlowVolatility$⁠). In case that fewer than five years of lagged data are available, the standard deviation is calculated over all available years, but is set to missing if fewer than 3 years are available (see Graham and Leary, 2018). Additionally, we control for firms with blockholders possessing at least 25% of a firm’s voting rights but not being a founding family firm (⁠$OtherUltimateOwner$⁠). The variable is defined as a dummy variable equal to one if a firm has an $OtherUltimateOwner$ in a particular year and zero otherwise. All continuous variables are winsorized at the 1% and 99% levels to remove the influence of outlier observations.

Finally, to account for macroeconomic conditions that might influence cash holdings and at the same time be responsible for changes in labor regulations, all regressions include the following country-level variables: the inflation rate (Inflation) to control for the opportunity costs of holding cash; the growth of the gross domestic product (⁠$GDPGrowth$⁠) and the natural logarithm of GDP per capita (⁠$ln(GDPpc.)$⁠) to proxy for aggregate investment opportunities (Duong et al., 2020; Graham and Leary, 2018). Table A1 in the  Appendix provides detailed descriptions of all variables used in this study.

Panel A of Table 1 reports summary statistics for the full sample. The average cash holdings (Cash) in our sample are approximately 15%, and 25% when measured in terms of CashToNetAssets. These numbers are fairly close to those of Karpuz et al. (2020).

The average sample firm has 3,215 million EUR in total assets, with NWC accounting for 1.8% of total assets. The average firm trades at a market-to-book ratio of 1.9, has debt that makes up for 20.5% of total capital, invests 8.6% of total assets per year, has a positive cash flow, which amounts to 2.9% of total assets, with a volatility of 9.4% as the average over all industries. More than half of all firms (61.3%) in the sample pay dividends and 31.9% have other ultimate owners. 22.1% of all firm-year observations in the sample refer to founding family firms (FFF).

The LMR index has a mean value of 4.011 and a standard deviation of 1.361, revealing substantial cross-country variation. Figure 1 shows the staggered character of LMR changes and variation within countries. In total there are 74 increases and 96 decreases in LMR that are fairly evenly distributed across countries and through the sample time. The average country has an Inflation rate of 1.5%, a $GDPgrowth$ of 0.9% and a GDP p.c. amounting for 44,606U.S. Dollar.

This section presents our empirical results. Section 3.1 begins with simple t-tests before introducing our baseline regression results. Section 3.2 examines cross-sectional heterogeneity, while Section 3.3 discusses the economic significance of our findings. Robustness checks are provided in the Online Appendix.

To study the role of FFFs in the relationship between firms’ cash holdings and labor market regulations, we proceed in two steps.

First, we use simple t-test to compare cash holdings of FFFs and non-FFFs in different subsamples, depending on the level of labor market regulation in the firm’s country of domicile. Results of this exercise are reported in Table 2.

Interestingly, there is not too much of a difference between FFFs and non-FFFs cash holdings for the full sample. Indeed, in terms of Cash, the difference is statistically insignificant. In terms of CashToNetAssets, the difference is significant on the 5% level, but fairly small with 1.6 percentage points.

Looking at the differences in means between FFFs and non-FFFs in Table 2, we see that for low LMR country-year observations the difference is −0.006 for Cash and −0.056 for CashToNetAssets and thus lower than the difference in the full sample (+0.002 for Cash and −0.016 for CashToNetAssets) and vice versa for high LMR country-year observations ^[13]. Moreover, while for low LMR country-year observations the differences are negative, they are positive for high LMR country-year observations (+0.019 for Cash and +0.046 for CashToNetAssets). The underlying mechanism for the pattern is that non-FFF seem to react more cautiously to strengthening LMR than FFF do. Both observations support our hypothesis that family firms will increase their cash holdings in response to tightening labor regulations relative to other firms.

Second, we extend the regression model adopted by prior literature (see e.g. Cui et al., 2018; Klasa et al., 2009) for an interaction term between FFF and LMR as follows:

where $Cashit$ measures cash holdings of firm i at time t, $LMRct$ the extent of labor market regulation in country c (the country of domicile of firm i) at time t, $FFFit$ indicates the FFF status of firm i at time t, and $Controlscit−1$ refers to control variables on (firm- and country-level) as discussed in section 2.2.4. Moreover, $θcjt$ denotes the set of fixed effects and $εit$ the error term. When estimating Model (1) we allow for:

• firm-fixed effects to control for unobserved time-invariant firm characteristics;

• industry-year fixed effects (based on 48-Fama-French industry classification) to control for possible unobserved time-varying industry shocks that might affect firms’ cash holdings (e.g., Chen et al., 2018; Gao et al., 2013; Harford et al., 2008);

• country-specific year trends to control for linear developments that might coincide with changes in LMR; and

• standard errors that are robust to within-country correlation of residuals, as LMR changes vary at the country level ^[14].

Our variables of interest in Model (1) are $LMRct$ referring to the extent of labor market regulation in country c, and, more importantly, its interaction with $FFFit$⁠, the dummy variable indicating the FFF status of firm i at time t. As such, we are interested in $β1$ and, more specifically, in $β2$⁠, and our prediction is $β2>0$⁠.

Table 3 reports estimates of Model (1) with and without the interaction term, for Cash (Columns 1 and 2) and CashToNetAssets (Columns 3 and 4). Before looking at the coefficients of interest, we note that all firm-level control variables have signs predicted by the literature. Size, Leverage, NWC, and Investment are negatively related to cash holdings and statistically distinguishable from zero. The coefficients on Growth, $DividendDummy$⁠, $CashFlow$⁠, $CashFlowVolatility$⁠, and $OtherUltimateOwner$ show a positive sign, whereby only the coefficient on Growth is statistically significant through all specifications. The coefficients on the country-level controls are mostly statistically insignificant.

Columns 1 and 3 show the regression results for the reduced form of Model (1), i.e. excluding the interaction term $LMRct×FFFit$⁠. For both definitions of cash holdings, the coefficient estimate for LMR has a negative sign, consistent with the “bargaining power view”. However, in both cases the coefficients are statistically insignificant at conventional levels.

Columns 2 and 4 present estimation results of Model (1) with interaction between LMR and the FFF status. Two observations stand out. First, consistent with the “bargaining power view”, the estimated coefficient for $LMRct$⁠, i.e. the LMR-sensitivity estimate for non-FFF, is negative and statistically significant at the 5% level for both definitions of cash holdings. Second, the estimated coefficients for the interaction term $LMR×FFF$ are positive. Also, they are statistically significant at the 1% level in Column 2 and at the 5% level in Column 4. The latter observation supports our hypothesis that FFFs will increase their cash holdings in response to tightening labor market regulation relative to non-FFFs, and provides suggestive evidence for the idea that that “financial flexibility” considerations might be more important for FFFs when deciding about their response to labor market regulation, while “bargaining power” arguments might be more relevant for other firms.

This section examines cross-sectional heterogeneity in the previously documented FFF-specific response to LMR to strengthen our arguments. In Section 3.2.1, we examine the role of labor costs, aiming to strengthen the argument that the observed pattern is caused by variations in LMR. In Section 3.2.2, we examine the role of operating risk to strengthen the argument that risk mitigation arguments and a precautionary demand for cash holdings in FFFs after tightening LMR can help to explain the observed pattern.

We propose two test to strengthen our argument that the observed pattern is caused by variations in LMR. Both build on the idea that labor regulation alters labor costs, and thus the extent of labor costs exposed to changes in LMR should matter for the response to LMR changes.

First, we argue that labor regulation should be particularly relevant for labor intense firms, where labor represent an important production factor. To test this prediction, we split the sample into two subsamples, depending on whether a firm’s labor intensity is above (equal or below) the median in the affiliate industry in a particular year and estimate Model (1) on each of the subsamples separately. To classify firms, we follow Hilary (2006) and calculate firm’s labor intensity as the number of employees divided by total assets.

Second, we argue that variations in domestic LMR should be less relevant for multinational firms, as these firms have shifted (or can shift) operations to unaffected countries and thus ‘escape’ tighter labor market regulations in the home country. To test this prediction, we again split the sample into two subsamples: domestic firms versus multinationals. We classify firms as domestic firm, if the firm does not report any foreign assets in the previous three years ^[15], and as multinational otherwise.

We re-estimate the Model (1) for all sub-samples and present results in Table 4. A first look at the table confirms that while in all specifications our coefficients of interest retain their signs, they vary in size and significance. Moreover, this variation is systematic and consistent with the underlying idea of this exercise: The response to variations in LMR is more salient in labor-intensive firms (compared to other firms) and in domestic firms (compared to multinational firms).

Interested in the FFF-specific response to LMR changes, we then need to look into the differential response across subsamples. Column 1 and 2, as well as 3 and 4, contrast results for the subsample of labor-intense firms versus results for the subsample of other firms. We find the FFF-specific response to LMR changes for labor-intense firms to be almost twice as high in magnitude compared to the average response (i.e. full sample analysis in Table 3) and some five to seven times the response in other firms. These differences also have consequences for the statistical significance of the coefficient estimates on the different subsamples. Column 5 and 6, as well as 7 and 8, contrast results for the subsample of domestic firms versus results for the subsample of multinational firms. We find the FFF-specific response to LMR changes for domestic firms to be three to four times the average response (i.e. full sample analysis in Table 3) and some four to seven times the response in multinational firms. Overall, while we do not test for statistical significance in the differential response (across the different subsample), the results support our notion that the pattern observed in Table 3 is caused by variations in LMR.

Risk-aversion and the tendency to engage in risk mitigation is one of the underlying arguments for our hypothesis of an FFF-specific response to variations in LMR. Specifically, we argue that FFFs react with a precautionary demand for cash holdings in response to strengthened LMR, as costs of financial distress are elevated by tighter labor regulation. Following this logic, the impact of tighter LMR should be more pronounced for FFFs with riskier business models, leading them to react more strongly to a change in LMR.

With this idea in mind, we examine whether FFFs that run more risky business models show a stronger response to variations in LMR. We gauge the riskiness of a business model using to proxies. First, we follow Kim and Zhang (2014) and use variations in sales as a proxy for operating risk. We calculate $SalesVolatility$ as the standard deviation of sales, deflated by total assets, in the previous five years (Wei et al., 2020), with at least three out of five available firm-year-observations. Second, we follow Karpuz et al. (2020) and proxy firm risk using $CashFlowVolatility$⁠.

Next, we split the sample with respect to the FFF status and estimate the following model on each of the subsamples:

where $Riskit$ denotes sales or cash flow volatility, while all other variables follow the same notation as in Model (1). Table 5 presents the results of this exercise.

Columns 1, 3, 5, and 7 present results for the subsample of FFF. In line with the risk mitigation argument, in each Column the coefficient for the interaction term between LMR and Risk is positive and significant at the 1% level, suggesting that FFFs that are exposed to more risk respond with a higher demand for cash holdings to rising LMR^[16]. As such, Table 5 provides further supports for our theoretical argument, that risk mitigation intentions represent an important factor for the FFF-specific response to variations in LMR.

So far, we have only discussed statistical significance of our results. To assess the economic significance of our findings, we use the point estimates of the regression parameters from our baseline findings (Table 3) as well as from the results on labor-intensive and domestic firms (Table 4) and calculate the elasticity measure as dx/dy$×$(Ø[x]/Ø[y]) for both definitions of cash and with respect to all explanatory variables. Table 6, Columns 1 and 2 show the economic significance using the full sample, Columns 3 and 4 for labor-intensive firms, and Columns 5 and 6 for domestic firms ^[17] on Cash and CashToNetAssets respectively.

Based on the universe of all firms in our sample, Column 1 (2) shows, that while a 1% increase in LMR produces a 0.110% (0.257%) decrease in Cash (CashToNetAssets) for non-$FFF$⁠, an average FFF reacts with 0.083% (0.129%) increase in its Cash (CashToNetAssets) position. Thereby the average heterogeneous effect between the two groups amounts for 0.193% (0.386%). As expected, the economic effect is more pronounced in firms that largely rely on labor or cannot shift their production to other, currently less regulated countries. As shown in Column 3 (4), a 1% increase in LMR results in 0.202% (0.560%) decline in Cash (CashToNetAssets) for labor-intensive non-FFF, but leads to 0.202% (0.243%) raise in Cash (CashToNetAssets) for labor-intensive FFF. This corresponds to an average heterogeneous effect of 0.404% (0.802%). With regards to domestic firms, Column 5 (6) shows that on average Cash (CashToNetAssets) diminishes by 0.388% (0.743%) for each 1% increase in LMR for non-FFF, but rises by 0.155% (0.405%) for FFF. This corresponds to a 0.543% (1.148%) difference in the cash holdings’ response of FFF compared to non-FFF.

In comparison, other cash holding determinants have a much lower economic effect in our sample. Only firms’ Size and $ln(GDPpc.$⁠) show stronger elasticity to cash holdings, albeit $ln(GDPpc.)$ is not statistically significant in most of the specifications. The economic effect of Leverage on cash holdings is comparable to those of LMR in the full sample, but is considerably lower for labor-intensive or domestic firms.

The results in Table 6 indicate that:

• the FFF status has an economically significant impact on the sensitivity of cash holdings’ response to changes in LMR; and that

• LMR appears to be an economically important determinant of cash holdings policy in both FFF and non-FFF.

This section addresses two key empirical challenges. First, one might argue that FFF status is an endogenous choice influenced by the level of LMR. We tackle this concern in Section 4.1 by employing an instrumental variable (IV) regression, instrumenting for FFF status. Second, FFFs and non-FFFs may systematically differ, raising the possibility that our findings reflect underlying heterogeneity rather than a true causal relationship. To address this, we apply propensity score matching (PSM) in Section 4.2- and conduct a series of horserace regressions in Section 4.3.

A family’s decision to maintain or dilute its ownership–and thus the FFF status of the firm–may depend on its financial policy and associated constraints, as well as LMR and other aspects of the contracting environment. Families’ decision to maintain or to dilute their ownership and thus the FFF status might be a function of LMR, firms’ financial policy, financial constraints and other characteristics of firm’s contracting environment. As such, one might argue that the observed pattern may suffer from reverse causality and self-selection issues. While we cannot completely rule out these concerns, we try to mitigate them by using a two-stage least square instrumental variable approach.

To instrument the FFF status, we consider three country-level variables ^[18]. First, we follow Bennedsen et al. (2019) and use respondents’ assessments of the strength of family values in a particular country from the World Value Survey, based on the argument that high family values are associated with a higher probability to maintain family ownership. We define the measure of family value as the mean response on the country-year level using data from the year that is closest to the beginning of our sample (2007) ^[19]. We construct a dummy variable (⁠$HighFamilyValue$⁠) equal to one if the mean response in a particular country is above the median value of the sample. Second, we consider the country score on the individualism dimension of Hofstede’s model of national culture (Individualism). We expect that family firms are more common in countries with low individualism scores, indicating that individuals prefer to be part of strongly tied groups that shape their preferences and interests (Hofstede, 2001). Finally, we use the country-level data on inheritance law from Ellul et al. (2010) as a measure of legal context that is evident to impact family decisions on intergenerational transfer (see Ellul et al., 2010). We construct the $InheritanceLawPermissiveness$ variable as specified in Section.

Table 7 reports results of the first-stage estimation of the FFF dummy in Column 1 and the estimation of the interaction of FFF dummy and LMR index in Column 2. As instruments, we use country-level variables, as described above, and interaction terms of each of these variables with LMR. The first and second stages also include all other control variables that enter Model (1), except country-specific year trend and firm fixed effects since our instruments are on the country level and time-invariant. As predicted, the FFF status is positively correlated with $HighFamilyValue$ and $InheritanceLawPermissiveness$ negatively correlated with Individualism. Columns 3 and 4 of Table 7 show results on the second-stage estimation for Cash and CashToNetAssets respectively. For both definitions of cash holdings, our results continue to provide strong evidence of the negative relationship between changes in LMR and cash holdings in non-FFFs and a positive one in FFFs. As expected, the magnitude of the coefficients of interests in the second stage is higher than that reported in Table 3, while the significance remains largely unchanged. Further, the first-stage Kleibergen-Paap Wald F-tests and second-stage Hansen’s J test suggest that the relevance and exogeneity conditions are verified for our instruments.

Overall, we conclude that our baseline results are robust to concerns related to the endogeneity of the FFF status.

The literature documents that family firms differ from non-family firms, for example regarding size (e.g., Anderson and Reeb, 2003), leverage (e.g., Schmid, 2013), investments (Anderson et al., 2012) or age (e.g., Franks et al., 2012). Hence, one could argue that the presented results are simply driven by heterogeneous firm characteristics and not, as assumed, by a differing set of incentives specific to FFFs. Examining this concern, we form a matched sample by assigning non-FFFs with similar firm characteristics as a control group to the FFF observations, by using a propensity score matching (PSM) approach to minimize possible heterogeneity between these two groups.

We match non-FFF observations to FFF-observations within country ^[20] and industry ^[21] by a set of firm-level matching variables using nearest neighbor matching without replacement ^[22]. The matching variable firm size, investments, and leverage following Bennedsen et al. (2019), and firm age, as an additional discriminant factor that has been discussed in the literature (e.g., Franks et al., 2012).

Table 8 presents the results of the PSM approach without (Columns 1 and 3) and with a caliper restriction of $δ=0.0001$ (Columns 2 and 4) for Cash and CashToNetAssets as dependent variables. Showing quantitatively and qualitatively similar results in all specifications, our main findings appear to be robust against the concern of firm specific heterogeneity between FFFs and non-FFFs.

One might argue that our results could be explained by two alternative explanations on firm-level. First, one could argue that high-tech firms and financially constrained firms have a higher need for financial flexibility and thus might increase their cash holdings as a response to tightening LMR (e.g., Karpuz et al., 2020; Schmalz, 2016). At the same time, high-tech status and financial constraints might arguably correlate with FFF status. We address this concern by running a number of horse race regressions, where we add additional variables interacted with FFF status to Model (1). Results of this exercise are reported in Table 9.