Sensitivity to shocks and implicit employment protection in family firms

Accepted Manuscript Title: Sensitivity to Shocks and Implicit Employment Protection in Family Firms Author: Carl Magnus Bjuggren PII: DOI: Reference:

S0167-2681(15)00202-4 http://dx.doi.org/doi:10.1016/j.jebo.2015.07.011 JEBO 3633

To appear in:

Journal

Received date: Revised date: Accepted date:

29-10-2014 12-7-2015 20-7-2015

of

Economic

Behavior

&

Organization

Please cite this article as: Carl Magnus Bjuggren, Sensitivity to Shocks and Implicit Employment Protection in Family Firms, Journal of Economic Behavior and Organization (2015), http://dx.doi.org/10.1016/j.jebo.2015.07.011 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

*Title Page (with Full Author Details)

Research Institute of Industrial Economics (IFN)

us

a

a

cr

Carl Magnus Bjuggren

ip t

Sensitivity to Shocks and Implicit Employment Protection in Family Firms

an

Corresponding author Carl Magnus Bjuggren

M

Research Institute of Industrial Economics (IFN) Box 55665, SE-102 15, Sweden

Ac ce p

te

d

Email: [email protected] Phone: +46 8 665 45 82

Page 1 of 53

*Manuscript

Abstract

ip t

In this study I find that employment in family firms is less sensitive to performance and product market fluctuations. I show this by investigating aggregate fluctuations at the industry level as well as idiosyncratic firm level shocks. By differentiating between temporary and permanent shocks at the firm level, I find that family firms appear to be less anxious to translate temporary shocks into changes in employment. This supports the idea that family

us

cr

firms are able to offer their employees implicit employment protection. Family firms are believed to have longer time horizons, and are as owners more easily identified with their company and its actions. These are features that could make family firms more cautious

an

in terms of adjusting their employment. Unlike previous contributions, I am able to identify all family firms, both private and public, by using full population data from tax registers.

te

d

M

Keywords: Family Firms, Employment Protection, Shocks JEL classification: G32, J23, L25, D22, J21

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Preprint submitted to Journal of Economic Behavior & Organization

July 11, 2015 Page 2 of 53

1. Introduction In most countries, family firms constitute the majority of privately held firms and a significant part of publicly held firms (Astrachan and Shanker, 2003; Bertrand and Schoar,

ip t

2006; Faccio and Lang, 2002; La Porta et al., 1999; Morck et al., 2005). Consequently, scholars are showing an increased interest in the performance and behavior of family firms in relation to firms with more dispersed ownership. Several studies seem to agree upon

cr

the idea that family firms have longer time horizons and are as owners and managers more easily identified with the company and its actions (Anderson and Reeb, 2003; Block, 2010; Chandler and Hikino, 1990; Landes, 1949; Sraer and Thesmar, 2007). These are features that

us

could affect the firm’s employment decisions. Recent studies indicate that family firms are more cautious in their hiring and firing decisions (Bassanini et al., 2013; Sraer and Thesmar,

an

2007). In this paper, I present empirical results that show that employment in family firms is less sensitive to product market and performance fluctuations. This supports the idea that family

M

firms are able to offer their employees implicit employment protection. This also implies that family firms can potentially function to smooth out employment over business cycles. I find

d

that employment in family firms is less sensitive to sales and value added fluctuations at the industry level. However, publicly listed family firms do not show a similar sensitivity. I further extend the analysis to show that family firms are less sensitive to unanticipated

te

shocks by filtering out the trend component. I then proceed with investigating idiosyncratic shocks at the firm level and my results suggest that family firms are less anxious to translate temporary shocks in performance and sales into changes in employment. This study adds to a small but growing literature that focuses on the difference between family and non-family firms when it comes to employment adjustments (Bassanini et al.,

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

2013; Block, 2010; Lee, 2006; Sraer and Thesmar, 2007; Stavrou et al., 2007). The study by Sraer and Thesmar (2007) is the only one to investigate the impact of shocks to production related measures. The lack of work in this area is due primarily to the lack of firm ownership data. I add to the current literature by utilizing a novel way of identifying family firms, the methodology of which has been previously described in Bjuggren et al. (2011). By using tax register data, I am able to analyze all family firms in the Swedish economy, both private and public.1 The identification is made possible by a tax reform in the early 1990s that required the Swedish Tax Authority to identify family relations among ultimate owners of every Swedish firm. 1

I will throughout this paper refer to publicly traded firms as public and private firms that do not offer to trade the company to the public as private.

1 Page 3 of 53

The analysis of employment adjustments in family firms has typically been confined to rather crude estimations. For example, the inclusion of fixed effects have resulted in either analysis of ownership transfers or forced the level of analysis to the industry rather than to the individual firm. When estimating the effects of industry sales or value added fluctuations, one will capture movements that are due to both a trend component and a random

cr

ip t

component. In contrast to Sraer and Thesmar (2007), I filter out the trend component in a first stage regression on sales and value added and use the remaining residual as a measure of unanticipated shocks at the industry level. Furthermore, using variation at the industry level

us

could be deceptive if family firms are systematically less exposed to shocks. I use a method proposed by Guiso et al. (2005) which enables me to analyze employment adjustments at

an

the firm level taking unobserved time invariant characteristics into consideration as well as allowing me to distinguish between temporary and permanent shocks to the firm. A more recent adoption of this strategy include Cardoso and Portela (2009) and K´atay (2008).2 I begin by summarizing the previous literature on family firms and employment adjustments. Section 3 describes the data and presents some descriptive statistics on family firms.

M

Section 4 presents the empirical estimations that investigate the sensitivity of employment to industry and firm level shocks. Section 5 concludes the paper.

d

2. Theoretical Considerations and Previous Literature

te

An essential problem is to decide who shall bear the risk associated with exogenous fluctuations in production. Stiglitz (1987) argues that letting the worker bear the risk of these fluctuations is inefficient. Family firms could possibly solve some of the commitment problems involved with insuring employees against this risk. Frequent changes in ownership

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

structure could create a situation where the new owners or managers do not have to honor previous contracts. Shleifer and Summers (1988) discuss the same problem and stress the importance of loyalty and long-term commitment. They argue that a person that spends a long time within a company before becoming a CEO will develop a commitment to the stakeholders. In particular, Shleifer and Summers (1988) mention that the offspring of family firms could be raised with a certain loyalty towards all parties involved in the business. The idea that family firms could be more able to commit to implicit contracts with their employees has been investigated empirically by Bach and Serrano-Velarde (2015), Bassanini et al. (2013), and Sraer and Thesmar (2007). All three studies are using data on French firms and find support for the idea that family firms are offering implicit employment insurance 2

The outcome variable in these studies is wage instead of employment. There is a working paper by Ellul et al. (2014) that also investigates family firm’s sensitivity to both temporary and permanent shocks. Their paper is inspired by Guiso et al. (2005) but they use a slightly different specification.

2 Page 4 of 53

in return for lower wages. Additional studies by Block and Spiegel (2011), Lee (2006) and Stavrou et al. (2007) analyze downsizing and employment adjustments in family firms. All three studies investigate family firm behavior among public firms in the US. Lee (2006) and Stavrou et al. (2007) present descriptive findings that indicate that family firms are associated with less downsizing and appear to maintain relative employment stability during

cr

ip t

temporary market downturns. Block (2010) finds that family firms are less likely to downsize when it comes to deep job cuts.3 The feature of risk aversion in family firms is often mentioned in the literature (Bandiera

us

et al., 2011; Chandler and Hikino, 1990; Landes, 1949; Morck, 2000; Sraer and Thesmar, 2007), but the implications on employment adjustments have not been discussed in detail.

an

In fact, risk averse firms, regardless of ownership, are believed to pursue a faster adjustment of employment back to steady state when hit by a sudden shock. Risk aversion serves as a self-adjusting device in the sense that it steers the firm closer to the steady state level of profits, minimizing fluctuations (Choudhary and Levine, 2010). An interesting branch of literature within the field of business and management is challenging the notion of general

M

risk aversion in family firms. The argument is based on the risk of losing what is called ”socioemotional wealth”, which refers to an emotional value on non-financial aspects of the firm. Socioemotional wealth includes, for example, the ability to exercise family influence as

te

d

well as the preservation of the firm. The ability to exercise influence and the survival of the firm then becomes an end in itself. A family firm is believed to be more willing to accept, for example, below-target performance to avoid the loss of socioemotional wealth. Family firms are also believed to avoid venturing risk, which includes search for new technology and products (Berrone et al., 2012; G´omez-Meja et al., 2007). This is in line with the results

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

that family firms appear to be less anxious to translate temporary shocks into employment adjustments at the same time as they do not seem to differ in their response to permanent shocks. The preservation of socioemotional wealth is believed to have an effect on employment practices. Non-family employees are thought to acknowledge that family employees are being favored without proper regard to their qualifications. Family firms are believed to be more likely to attract employees that are motivated by factors such as lifetime job security and informal work environment rather than economic rewards and other perquisites that are typically reserved for family members (Cruz et al., 2011). The close tie between a family owner, or manager, and their company might, for example, 3

In addition there are two working papers that relate to this topic. Bach (2010) shows that family firms show less volatile sales and employment growth paths. D’Aurizio and Romano (2013) show that in the aftermath of the 2008 crisis, family firms chose to safeguard workplaces close to the firm’s headquarters.

3 Page 5 of 53

be manifested in the company name. This makes them more easily sanctioned and monitored by society. The idea that family owners and managers are more responsible towards their employees is put forward by Dyer and Whetten (2006), Stavrou et al. (2007), and Wiklund (2006). There is also a branch of literature that argues that family firms are guided to a large extent by altruistic incentives and act like stewards rather than agents (Miller and

cr

ip t

Le Breton-Miller, 2006). Finally, inclusion of both public and private firms in the empirical analysis avoids potential selection bias as firms that choose to enlist in the stock exchange may have different

us

growth ambitions and need for finance. As shown by Davis et al. (2007), growth rate variability in public and private firms follow different trends over time. 3. Data

an

The data used are register data from Statistics Sweden (SCB) on all firms with at least five employees, and covers the period from 1997 to 2009.4 There is little agreement on

M

how to define a family firm. Practice ranges from more broad definitions, such as a firm in which a family is in control over the strategic decisions of the business, to more narrow

d

definitions, such as a firm in which the founder and multiple generations are active and have declared the desire to keep the business within the family (Astrachan and Shanker, 2003). For example, the European Commission (2009) lists more than 90 definitions, most of which

te

are not applicable to studies using large data sets. In this paper, I will apply the definition of a family firm as a firm in which a family or a single individual controls more than 50 percent of the ultimate voting rights. This definition is slightly stricter than the one used by Faccio and Lang (2002) and La Porta et al. (1999), where a family or a single individual has to control more than 20 percent of the voting rights.5

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

An alternative to using concentrations of ultimate voting rights would be to look at family CEO transitions (Bach, 2010). This would incorporate the idea that the firm is intended to be kept within the family. One advantage of using ultimate voting rights as identification is that it accounts for the case where a firm appoints an outside CEO while keeping indirect control of the firm through the majority of voting shares. If a CEO does not operate the way shareholders want her to, they can simply replace her with another. There are, however, potential problems with using ultimate voting rights as a basis for the definition of family firms. Some of the theories presented in the previous section relied upon the fact that family firms were passed down through generations or that the owner and the manager is the same 4 5

See online Appendix B for details on data construction. I will use the terms family firm and family-owned firm interchangeably.

4 Page 6 of 53

person. A drawback with this study is that the data do not allow for the distinction between family managed versus professionally managed firms. However, for firms in which family ownership exceeds 50 percent, the majority are likely to be family managed (Bjuggren et al., 2011). As of 1993, the Swedish Tax Authority has identified the number of owners in all private

cr

ip t

closely held firms. With this identifier I am able to trace all private incorporated firms in which a family or an individual control at least 50 of the firm.6 To identify publicly held firms, I used the standard work on ownership in public firms in Sweden, Owners and Power

us

in Sweden’s Listed Companies by Sundqvist (1997-2009). The data are restricted to incorporated firms, and in order to reduce systematic size

an

differences between family and non-family firms, firms with less than 500 employees are excluded.7 There is little variation in ownership concentration for larger firms. In total, only about 6 percent of the firms with more than 500 employees are identified as family firms. Moreover, the sample is restricted to exclude industries such as fishing, forestry, and agriculture. A complete list of the included industries and the distribution of family

M

firms within each industry, as well as the distribution of sales and value added within each industry, can be found in the online Appendix B.8 Both sales and value added appear to fluctuate more in industries such as financial intermediation, air transport and recycling,

te

d

whereas there is less fluctuation in industries such as publishing, real estate and some types of manufacturing. Table 1 shows that about 44 percent of companies are family firms and they contribute to 29 percent of the total employment. These figures are in line with previous research on the prevalence of family firms (Anderson and Reeb, 2003; Bach, 2010; Bjuggren et al., 2011;

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Faccio and Lang, 2002; La Porta et al., 1999). The comparability with countries such as the US and France strengthens the argument of external validity. In addition, family firms correspond to 25 percent of public firms and account for about 30 percent of employment within this sub-sample.9 Looking at differences between family and non-family firms, Table 2 reveals that family firms are on average smaller, and sales and value added is lower. The main differences appear to be located in the top 75 percentiles. Moreover, family-owned 6

See Appendix A for details on the construction of the family firm variable. The results do not change with this restriction. All estimations are reproduced including firms with more than 500 employees. See Tables B1-B4 in the online Appendix B. The size restrictions do not apply to the sample of public firms used in Table 3. I use the term incorporated firm to describe the Swedish legal form aktiebolag, which refers to a stock corporation, sometimes also translated as limited company. 8 See Tables B5 and B6. 9 All estimations are reproduced using the sub-sample of private firms. See Tables B7-B10 in the online Appendix B. The results are almost identical since private firms vastly outnumber public firms. 7

5 Page 7 of 53

firms are slightly older on average. The size distribution of family and non-family firms is shown in Figure 1. There are relatively more family firms in the smaller size categories. As a point of reference, there are in total 59,752 family firms with five employees and the corresponding figure for non-family firms is 52,904. The geographical distribution of firms is shown in Figure 2. By using data

cr

ip t

on establishments, I am able to locate firms according to municipality. The left map shows the distribution of the absolute number of firms over municipalities, 1997-2009. It becomes evident that most firms are located around the larger metropolitan areas: Malm¨o in the

us

south, Gothenburg (G¨oteborg) on the west coast, and Stockholm on the east coast. The right map shows the distribution of the share of family firms in each municipality. The distribution reveals that family firms are relatively less frequent in the three large metropolitan areas.

an

4. Empirical Estimation

In order to estimate a causal effect of family ownership on employment adjustments,

M

random allocation of firm ownership would be required. The absence of such a randomized experiment raises some challenges. First, if family firms are systematically less exposed to

d

shocks there may be bias in estimations at the industry level. The subsequent section 4.3 that focuses on idiosyncratic shocks at the firm level will mitigate this potential bias. Second, as shown in Table 2, family firms are smaller in general. If small firms are also less susceptible

te

to market fluctuations, we would have a potential bias where the family firm indicator would capture effects that are due to size rather than ownership structure. The inclusion of firm fixed effects should capture characteristics such as size if it is constant over time. A potential bias could come from a correlation between firm growth and ownership structure. Growth ambitions could be correlated with a choice of more dispersed ownership. If

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

growth ambitions are also associated with fluctuations in performance, we will have a competing mechanism that could serve as to explain the results. It is not within the scope of this study to try to separate dynastic motives from a lack of growth ambitions. A feature that is common to all family firms regardless of its motives is that the concentrated ownership makes them more easily sanctioned and monitored by society. Moreover, it is not obvious that observed differences in employment fluctuations are due to implicit employment protection. There could also in this case be a potential bias due to firm size. Smaller firms might be less able to provide insurance for their workers due to financial restrictions. This would however work against the results in this paper since family firms are found to be smaller in general. Another potential bias exists if there are differences in the level and impact of legislated employment protection. The level of employment protection in Sweden is arguably varying with size. It is possible for firms to 6 Page 8 of 53

evade some rules of employment protection through negotiating with the unions (Skedinger, 2010). Larger firms might have greater opportunities do so because of a more favorable bargaining position. Moreover, since 2001 employment protection is loosened for very small firms with less than 11 employees (Olsson, 2009; von Below and Thoursie, 2010). Hence, very small firms have a somewhat looser protection while the impact of employment protection for

cr

ip t

larger firms is likely to decrease with size. Like before, the firm fixed effects should capture features such as size if it is stable over time. The potential bias will then again come from firms that grow, if ownership concentration is endogenous to growth ambitions.

us

In the following empirical estimations, I will use panel data to control for covariates and fixed effects. The results are thus to be interpreted as associations rather than causal effects. 4.1. Industry Shocks

an

The methods in this section are aimed at measuring fluctuations in firm performance at the industry level to investigate how it affects employment in individual firms. Firm performance can be measured in a number of ways. I will use both sales and value added.

M

Cardoso and Portela (2009) argue that sales captures demand uncertainty since it directly reflects changes in product demand. Value added is according to Guiso et al. (2005) directly subject to stochastic fluctuations, and is not subject to discretionary reporting like profits.

d

In order to investigate family firm’s employment behavior, I model the sensitivity of employment to industry shocks based on the specification used by Sraer and Thesmar (2007).

te

They use fluctuations in total sales within an industry as their measure of shocks to firm performance. I will also add value added to the analysis. The following equation is estimated:

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

ln Yist = αis + β(Xist) ln Sist + γ(Xist )δt + υist

(1)

where ln Yist is the natural logarithm of employment in firm i, industry s, at time t, αis is firm fixed effects, ln Sist is the natural logarithm of total sales or value added in industry s, minus the contribution of the specific firm i, and δt is a year dummy capturing possible business downturns. β(Xist ) is to be interpreted as elasticity to industry shocks and γ(Xist ) as elasticity to economic wide shocks. These elasticities are supposed to vary across firms according to the following: β(Xist ) = β + bFist + c ln ageist + dSOEist + Γist e

(2)

γ(Xist ) = γ + b′ Fist + c′ ln ageist + d′ SOEist + Γist e′

(3)

where Fist is a family firm dummy, ln ageist is the natural logarithm of firm age, and

7 Page 9 of 53

SOEist is a dummy variable taking the value one if firm i has previously been state owned.10 Similar to France, Sweden has undergone a privatization of government-owned firms. Γist includes a dummy variable for being publicly listed, a dummy for belonging to a corporate group, and a dummy for operating in one of the greater metropolitan areas.11 The idea with this setting is to control for aggregate shocks and let firms vary in the response to industry-

cr

ip t

level fluctuations. Industry sales and value added might be correlated with the economy wide state δt and therefore, in the absence of equation (3), the estimates in equation (2) might be biased. This will be the case if, for example, family firms have overall experienced a

us

different growth rate than non-family firms during the time period 1997-2009. The equation (3) is thus controlling for aggregate shocks to the economy. In effect, putting the equations together, I am estimating the following:

an

ln Yist = αis + β ln Sist + bFist ln Sist + c ln ageist ln Sist + dSOEist ln Sist +Γist e × ln Sist + γδt + b′ Fist δt + c′ ln ageist δt + d′ SOEist δt + Γist e′ × δt + υist

(4)

M

where b measures the difference between family and non-family firms in the elasticity of employment to industry fluctuations . A negative estimated coefficient b would thus indicate that family firms respond less to sales or value added shocks within the industry. An implicit

te

d

assumption in this setting is that aggregate industry sales and value added are exogenous to sales and value added in the individual firm. This implies that each firm has to be small enough not to affect the total within each industry. The results from estimating the sensitivity of employment to industry shocks are presented in Table 3. All estimations include firm fixed effects, and columns (1) assume that

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

b′ = c′ = d′ = e′ = 0 in equation (3), i.e. that sales and value added shocks within industries are uncorrelated with economy wide shocks. Columns (2) relaxes the assumption and adds the variables in Γ to the equations. The coefficient of particular interest is the interaction between shocks and family ownership, ln Sist × family firm. With respect to all firms, the leftmost columns, the results reveal a negative estimated coefficient for the interaction between industry fluctuation and family ownership for all models. This indicates that employment in family firms is less sensitive to industry sales and value added shocks. In the right-most columns in Table 3, the same setting is applied to the sample of public 10

I am not able to differentiate between different types of CEOs as done by Sraer and Thesmar (2007). Note that these variables are not in the original specification by Sraer and Thesmar (2007). See Table B12 in the online Appendix B for an alternate specification without the dummies for corporate groups and metropolitan areas. 11

8 Page 10 of 53

firms.12 The most saturated model (2) that investigates sales shocks indicates a positive coefficient, but contrary to the previous results by Sraer and Thesmar (2007) there appears to be no persistent effect of family ownership. A possible explanation for the discrepancy is that Sraer and Thesmar (2007) are able to differentiate between different types of CEOs, labeling them as founder CEOs, heir CEOs, or professional CEOs. The effect could be offset

cr

ip t

when using an aggregate measure of ownership concentration. When analyzing all firms, as shown in the left-most columns in Table 3, the interaction between public firms and shocks becomes insignificant in the more saturated models. The private family firms appear to be

us

driving the results. This is also confirmed by Tables B7-B10, where I re-run the estimations for the sub-sample of private firms.13 4.2. Disentangling Industry Shocks

Sales and value added fluctuations at the industry level will capture both variation that

an

is explained by a trend as well as unanticipated shocks to this trend. In order to separate the random component I introduce a first stage equation that aims to filter out the trend.14

M

More specifically, I let the second and final equation take the same form as before, with the exception of the definition of the shock, here denoted by ǫist , (5)

d

ln Yist = αis + β(Xist)ǫist + γ(Xist )δt + υist .

or value added.

te

The unforeseen shock, ǫist , is the residual from a first stage predictive equation on sales

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

ln Sist = αis + β ln Sis,t−1 + Γist κ + δt + ǫist ,

(6)

where ln Sis,t−1 is the lagged logarithm of total sales or value added within an industry, αis are fixed effects, and Γist is defined as before with the inclusion of the natural logarithm of age. Since the fixed effects are correlated with Sis,t−1 the equation is estimated using the one-step first-differenced generalized method of moments (GMM) by Arellano and Bond (1991). In the GMM stage, higher levels of lags are used to instrument the one-period lagged logarithm of sales or value added.15 The GMM involves a step of first-differencing, which cancels out the fixed effects. The results from the GMM estimations are presented in Table 4, and the observed serial correlations of the first-differenced residuals confirms the use of 12

There are no restrictions on size for this sample. See online Appendix B. 14 I am at this stage not assuming any specific structure of the error term. 15 Sales is instrumented with lags at t−4 and value added is instrumented with 4-8 period lags. A regression on value added using lags at t − 4 as instruments instead will produce similar coefficients. However, the Hansen J-statistic will have a p-value of 0.6, which is large enough to be of concern. 13

9 Page 11 of 53

instruments lagged t-2 periods or earlier. The Hansen J-statistic indicates the validity of the instruments for the regressions on both sales and value added. The estimated coefficient on lagged sales, 0.63, represents the persistence of sales over time and the corresponding coefficient for lagged value added is slightly higher at 0.75.16 The residuals from Table 4 are then used in equation (5), and the results are presented

cr

ip t

in Table 5. The interaction between the residual and the family firms indicator, ǫist ×family firm, is still negative and statistically significant across all specifications. The coefficients indicate that employment in family firms is less sensitive to unanticipated fluctuations at

family firms following a different trend on average.

an

4.3. Firm level shocks

us

the industry level. This reinforces the idea that family firms are less sensitive to shocks, and indicates that the previous effect on total industry fluctuation was not merely a result of

Estimating responses to shocks within an industry could be misleading if, for example, family firms are systematically less exposed to shocks. Estimating a shock at the firm level

M

would mitigate this potential problem. In the following section, I will estimate firm level shocks using the specification proposed by Guiso et al. (2005).17 This entails estimating firm performance and employment separately, and using the residuals from these regressions to

d

capture idiosyncratic shocks. The residual from the employment estimation is regressed on the firm performance residual using different leads and lags as instruments to investigate

te

both permanent and temporary shocks. Similar to the previous setting, sales and value added are determined by

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

ln Sit = αi + ρ ln Si,t−1 + Xit θ + ǫit ,

(7)

where Si,t is the logarithm of sales in firm i at time t, Xit is a vector of controls including covariates from previous estimations, and αi is the firm specific effect.18 Since the fixed effects αi are correlated with Si,t−1 , OLS is inconsistent. Therefore, I will again use the generalized method of moments (GMM) by Arellano and Bond (1991). I report the results for the GMM in Table 6. Persistence of sales and value added is 16

For the sub-sample of public firms there is no observed persistence in sales or value added over time. See table B13 in the online Appendix B. It is therefore not feasible to separate the trend from the random component. 17 See Cardoso and Portela (2009) for an application on Portuguese data, and K´atay (2008) for an application on Hungarian data. 18 The covariates include the logarithm of firm age, a dummy variable for being publicly listed, a dummy variable for belonging to a corporate group, and a dummy variable for operating in one of the greater metropolitan areas.

10 Page 12 of 53

estimated to 0.87 and 0.38, respectively.19 The validity of the instruments used in the regression on value added are confirmed by the Hansen J-test. The first-differenced residuals show that the first and second lag are statistically significant, but the third lag is insignificant. This is similar to the results by Cardoso and Portela (2009), Guiso et al. (2005), and K´atay (2008). The regression on sales in Table 6 does not pass the test of over-identification. As a

cr

ip t

robustness check, all estimations are replicated in the online Appendix B using an alternate specification.20 Following Guiso et al. (2005), the error term from equation (7) can be decomposed

us

into a random walk and an MA(1) process. Using assumptions outlined in Appendix A, this decomposition is consistent with the pattern of significance for value added in Table

an

6, where the first-differenced residuals become statistically insignificant after two periods. The error term can can be written as the sum of a deterministic, a permanent, and a transitory component, the details of which are also described in Appendix A. The effect of

M

the permanent and the transitory component will later be estimated in an IV regression. I will assume that employment in a firm can be represented by the following equation ln Yit = Xit Φ + ϕi + αPit + βTit + ψit

(8)

d

where Yit is employment in firm i at time t, and Xit is a vector of controls defined as in

te

equation (7). I let the last four terms constitute the error term, representing fixed effects ϕi , permanent shocks Pi , temporary shocks Tit , and an error term ψit that is assumed to be uncorrelated with Pit and Tit . Employment is assumed to respond to permanent and temporary shocks with the sensitivities α and β respectively. I then multiply equation (8) by (1 − ρL), where L is a lag operator, and the process of employment takes the form

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Yit = ρYi,t−1 + (1 − ρL)Xit Φ + (1 − ρL)(ϕi + αPit + βTit + ψit )

(9)

and the error term can be defined as ωit = (1 − ρL)(ϕi + αPit + βTit + ψit ).

(10)

The transformation thus introduces state dependence on employment. Like before, I use the one-step first-differenced GMM to estimate the equation. The control variables in Xit are assumed to be exogenous. The results are presented in the right-most columns in Table 19

See Table 6 for information on the instruments used. The estimations are replicated in the online Appendix B using lagged sales in period t-7 and earlier as instruments (see Table B14 and B15). This results in a satisfactory Hansen J-statistic with a p-value of 0.38. However, the persistence of sales is then estimated to be 2.4, which might be of concern. 20

11 Page 13 of 53

6 and the estimated coefficient for employment at t − 1 is 0.97, which indicates a high degree of persistence. The test for over-identifying restrictions indicates that the instruments are valid. The assumed structure of the error term, ωit , in equation (10) is consistent with an MA(3) process but the first-differenced errors in Table 6 do not show any serial correlation in the third lag. Therefore, the exact structure of the error term, ωit , cannot be confirmed.

us

cr

ip t

Finally, the strategy proposed by Guiso et al. (2005) boils down to estimating α and β through regressing the residuals from the employment process ∆ωit on the residuals from the sales equation ∆ǫit . Guiso et al. (2005) show that α can be estimated based on the P instruments ( 2τ =−2 ∆ǫi,t+τ )k , and β can be estimated based on the instruments (∆ǫi,t+1 )k , for any k ≥ 1. These instruments are then interacted with the family firm dummy to capture

an

the heterogeneity in ownership concentration. Details on how the permanent and transitory components are captured are outlined in Appendix A. To estimate both α and β I have used the feasible efficient GMM procedure. Instruments are defined for k = 1, 2, 3 in each regression and the results are shown in Table 7. Estimations on temporary shocks indicate that family firms are less anxious to translate shocks in value

M

added and sales into employment changes. This is in line with the previous estimations on the industry level, suggesting that family firms are less sensitive to temporary shocks at the firm level.21 When I have instrumented ∆ǫit with the one period lead for k = 1, the

te

d

regressors are exactly identified. For k = 2 or k = 3, the estimations do not satisfy the over-identification test. The results are thus to be considered as tentative. The estimations trying to capture permanent shocks show no persistent significant effect of family ownership. The models are, like in the previous case, not confirmed by the over-identifying test which indicates that the equations are potentially miss-specified.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

The above results give some indications that family firms are behaving like any other firm when hit by permanent shocks, such as large technology changes, but are less inclined to make employment changes when hit by temporary shocks, such as machinery breakdown and temporary sales shocks. A rationale for this could be that when the future of the firm is at stake, which is more likely when hit by a permanent shock, family firms have at least as strong incentives to make sure that the firm stays alive. When hit by a temporary shock, the family firms’ features of long-term commitment, being more easily identified with the company and its actions, and willingness to accept below-target performance in order to avoid the loss of socioemotional wealth, could make them less anxious to adjust their workforce. 21

The analysis is not done separately for the sub-sample of public firms since the sample size is not large enough to support the first stage GMM estimations and the following feasible efficient GMM procedure, using both leads and lags as instruments.

12 Page 14 of 53

The results should be viewed as tentative since; i) the assumed structure of the error term from the employment equation is not verified by the observed serial correlation in the first-differenced errors, and ii) the iv estimations on permanent and temporary shocks are not validated by the Hansen J-test for over-identification.

ip t

5. Conclusions

cr

This paper shows that employment in family firms is less sensitive to product market and performance fluctuations. Family firms are believed to have longer time horizons and are as

us

managers and owners more easily identified with their company. These are features that could make family firms more cautious in terms of hiring and firing and thus enable them to offer their employees implicit employment protection. Sraer and Thesmar (2007) investigated how

an

publicly listed family firms react to shocks in performance, and their findings seem to confirm this hypothesis. Their analysis is, however, based on public firms where shocks are defined as sales fluctuations at the industry level. In this paper, I investigated how employment in

M

family firms responds to shocks in sales and value added, including both aggregate shocks at the industry level and idiosyncratic firm level shocks. I used full population data, which

d

allowed me to identify all family-owned firms in the economy, both public and private firms. As a first step, I confirmed the findings by Sraer and Thesmar (2007) and showed that family firms are less sensitive to industry sales shocks. I then extended the analysis and

te

found that it also holds for fluctuations in value added. However, contrary to the results found by Sraer and Thesmar (2007), public family firms in Sweden do not show a similar sensitivity to shocks. The sensitivity to industry sales and value added shocks in this paper are driven by the private family firms, which indicates that it is likely important to be able to differentiate between public and private firms.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Investigating sales and value added fluctuations at the industry level will capture both variation that is explained by a trend as well as unanticipated shocks to this trend. I proceeded by filtering out the trend in industry sales and value added fluctuations in order to isolate the unanticipated component. Employment in family firms is less sensitive to unanticipated shocks at the industry level as well. Using variation at the industry level could be misleading if family firms are systematically less exposed to shocks. This justifies the focus on idiosyncratic shocks at the firm level. In order to investigate how family firms respond to idiosyncratic shocks I relied on the framework by Guiso et al. (2005) and estimated the response to permanent and transitory shocks, respectively. The results should be considered as tentative but indicate that family firms are less sensitive to temporary demand shocks. Hence, family firms appear to be less anxious to translate temporary shocks in performance and product demand into changes in 13 Page 15 of 53

employment. In the absence of randomization of ownership concentration across firms, the presented results should be interpreted as associations rather than casual effects. Potential bias could come from a relationship between firm growth, volatility, and the choice of ownership structure. It is beyond the scope of this paper to try to separate dynastic motives from growth

cr

ip t

ambitions. Furthermore, it is not obvious that the observed differences in employment fluctuations are due solely to implicit employment protection. The impact of legislated employment protection might vary across firm size, which could also bias the results if ownership

us

concentration is endogenous to growth. The slower employment adjustments could be a result of family firms being inattentive.

an

The combination of ownership and management where, for example, a CEO is chosen based on kinship rather than talent, could lead to poor management. Large family firms have been shown to have weaker management than their non-family counterparts (Bloom and Van Reenen, 2010). The results presented in this paper could be driven by this feature, if weak management is systematically correlated with slower reactions to economic shocks. This

M

would still be consistent with an implicit employment protection being offered to workers who are well aware of the less responsive management in family firms. Moreover, if weak management is more easily detected in public firms and therefore prevented, it might explain

te

d

why public family firms do not show a similar sensitivity to shocks. Since the study of family firms and their employment behavior is relatively unexplored, I believe the findings are interesting regardless of them being a result of dynastic motives, a lack of growth ambitions, or weak management. The empirical results presented in this paper indicate that employment in family firms is less sensitive to performance and product

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

market fluctuations, which supports the idea that family firms offer their employees implicit employment protection. This implies that family firms can potentially function to smooth out employment over business cycles. The results also stresses the importance of modeling heterogeneity in firm ownership when it comes to risk-sharing and labor demand adjustments. For future work it would be interesting to extend the analysis to other countries that have different institutional frameworks and compare the outcomes. It is plausible that the level of legislated employment protection within a country has an effect on the sensitivity of employment to shocks. Sweden is a country with a relatively high degree of protection. The incentives to offer implicit employment protection might be higher in a country with a lower degree of legislated employment protection. Previous literature indicates that non-family employees might be willing to trade monetary rewards for increased job security. An interesting avenue for future research would be to investigate the selection of workers into family firms using data on individuals. Finally, being able to directly identify family involvement 14 Page 16 of 53

in the firm’s management practices would further expand research on the behavior of family firms. Acknowledgments

ip t

I am grateful for useful comments and suggestions from Sahaja Acharya, Ola Andersson, Per Hjertstrand, Dan Johansson, John Nye, Matilda Orth, Hans Sj¨ogren, Per Skedinger,

cr

Mikael Stenkula, and seminar participants at Stockholm School of Economics EHFF, CeFEO at J¨onk¨oping University, CESIS at KTH Royal Institute of Technology, and the Ratio

us

Institute. I gratefully acknowledge financial support from the Swedish Research Council for Health, Working Life and Welfare (Forte) grant number 2014–2740, the Jan Wallander and Tom Hedelius Research Foundation, the Marianne and Marcus Wallenberg Foundation, as

an

well as from Sparbankernas Forskningsstiftelse. References

M

Anderson, R. C., Reeb, D. M., 2003. Founding-family ownership and firm performance: evidence from the S&P 500. Journal of Finance 58 (3), 1301–1327.

d

Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58 (2), 277–97.

te

Astrachan, J. H., Shanker, M. C., 2003. Family businesses’ contribution to the U.S. economy: a closer look. Family Business Review 16 (3), 211–219. Bach, L., 2010. Why are family firms so small? Working Paper Series 3464, Paris December 2010 Finance Meeting EUROFIDAI - AFFI.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Bach, L., Serrano-Velarde, N., 2015. CEO identity and labor contracts: evidence from CEO transitions . Journal of Corporate Finance, in press. Bandiera, O., Prat, A., Guiso, L., Sadun, R., 2011. Matching firms, managers and incentives. NBER Working Papers 16691, National Bureau of Economic Research, Inc. Bassanini, A., Caroli, E., Reb´erioux, A., Breda, T., 2013. Working in family firms: less paid but more secure? Evidence from French matched employer-employee data. Industrial and Labor Relations Review 66 (2), 433–466. Berrone, P., Cruz, C., Gomez-Mejia, L. R., 2012. Socioemotional wealth in family firms: theoretical dimensions, assessment approaches, and agenda for future research. Family Business Review 25 (3), 258–279. Bertrand, M., Schoar, A., 2006. The role of family in family firms. Journal of Economic Perspectives 20 (2), 73–96.

15 Page 17 of 53

Bjuggren, C. M., Johansson, D., Sj¨ogren, H., 2011. A note on employment and gross domestic product in Swedish family-owned businesses: a descriptive analysis. Family Business Review 24 (4), 362–371.

ip t

Block, J., 2010. Family management, family ownership, and downsizing: evidence from S&P 500 firms. Family Business Review 23 (2), 109–130. Block, J., Spiegel, F., 2011. Family firms and regional innovation activity: evidence from the German Mittelstand. MPRA Paper 28604, University Library of Munich, Germany.

cr

Bloom, N., Van Reenen, J., 2010. Why do management practices differ across firms and countries? Journal of Economic Perspectives 24 (1), 203–24.

us

Cardoso, A. R., Portela, M., 2009. Micro foundations for wage flexibility: wage insurance at the firm level. Scandinavian Journal of Economics 111 (1), 29–50.

an

Chandler, A. D., Hikino, T., 1990. Scale and Scope: the Dynamics of Industrial Capitalism. Belknap Press, Cambridge, Mass.

M

Choudhary, M. A., Levine, P., 2010. Risk-averse firms and employment dynamics. Oxford Economic Papers 62 (3), 578–602.

te

d

Cruz, C., Firfiray, S., Gomez-Mejia, L. R., 2011. Socioemotional wealth and human resource management (HRM) in family-controlled firms. In: Joshi, A., Liao, H., Martocchio, J. J. (Eds.), Research in Personnel and Human Resources Management, Volume 30. Emerald Group Publishing Limited, pp. 159–217. D’Aurizio, L., Romano, L., Apr. 2013. Family firms and the Great Recession: out of sight, out of mind? Temi di discussione (Economic working papers) 905, Bank of Italy, Economic Research and International Relations Area.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Davis, S. J., Haltiwanger, J., Jarmin, R., Miranda, J., 2007. Volatility and dispersion in business growth rates: publicly traded versus privately held firms. In: NBER Macroeconomics Annual 2006, Volume 21. NBER Chapters. National Bureau of Economic Research, Inc, pp. 107–180. Dyer, W. G., Whetten, D. A., 2006. Family firms and social responsibility: preliminary evidence from the S&P 500. Entrepreneurship Theory and Practice 30 (6), 785–802. Ellul, A., Pagano, M., Schivardi, F., 2014. Employment and wage insurance within firms: worldwide evidence. EIEF Working Papers Series 1402, Einaudi Institute for Economics and Finance (EIEF). European Commission, 2009. Final report of the expert group. Overview of familybusiness-relevant issues: research, networks, policy measures and existing studies. [Accessed July 8, 2015] http://ec.europa.eu/enterprise/policies/sme/promotingentrepreneurship/family-business/family_business_expert_group_report_en. pdf. 16 Page 18 of 53

Faccio, M., Lang, L. H. P., 2002. The ultimate ownership of Western European corporations. Journal of Financial Economics 65 (3), 365–395.

ip t

G´omez-Meja, L. R., Haynes, K. T., N´ un ˜ ez Nickel, M., Jacobson, K. J. L., Moyano-Fuentes, J., 2007. Socioemotional wealth and business risks in family-controlled firms: evidence from Spanish olive oil Mills. Administrative Science Quarterly 52 (1), 106–137. Guiso, L., Pistaferri, L., Schivardi, F., 2005. Insurance within the firm. Journal of Political Economy 113 (5), 1054–1087.

cr

K´atay, G., 2008. Do firms provide wage insurance against shocks? Evidence from Hungary. Working Paper Series 0964, European Central Bank.

us

La Porta, R., Lopez-De-Silanes, F., Shleifer, A., 1999. Corporate ownership around the world. Journal of Finance 54 (2), 471–517.

an

Landes, D. S., 1949. French entrepreneurship and industrial growth in the nineteenth century. Journal of Economic History 9 (01), 45–61.

M

Lee, J., 2006. Family firm performance: further evidence. Family Business Review 19 (2), 103–114. Miller, D., Le Breton-Miller, I., 2006. Family governance and firm performance: agency, stewardship, and capabilities. Family Business Review 19 (1), 73–87.

d

Morck, R., 2000. Concentrated Corporate Ownership. University of Chicago Press, Chicago.

te

Morck, R., Wolfenzon, D., Yeung, B., 2005. Corporate governance, economic entrenchment, and growth. Journal of Economic Literature 43 (3), 655–720. Olsson, M., 2009. Employment protection and sickness absence. Labour Economics 16 (2), 208–214.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

SFS, 1999. Swedish Code of Statutes No. 1999: 1229. Inkomstskattelag [Income Tax Law]. [Accessed July 10, 2015] http://www.riksdagen.se/sv/Dokument-Lagar/Lagar/ Svenskforfattningssamling/Inkomstskattelag-19991229_sfs-1999-1229/. Shleifer, A., Summers, L. H., 1988. Breach of trust in hostile takeovers. In: Auerbach, A. J. (Ed.), Corporate Takeovers: Causes and Consequences. University of Chicago Press, Chicago, pp. 33–68. Skedinger, P., 2010. Employment Protection Legislation: Evolution, Effects, Winners and Losers. Edward Elgar, Cheltenham. Sraer, D., Thesmar, D., 2007. Performance and behavior of family firms: evidence from the French stock market. Journal of the European Economic Association 5 (4), 709–751. Statistics Sweden, 2009. Labour and education statistics 2009:1, integrated database for ¨ labour market research. Background facts, Statistics Sweden, Orebro. 17 Page 19 of 53

Stavrou, E., Kassinis, G., Filotheou, A., 2007. Downsizing and stakeholder orientation among the Fortune 500: does family ownership matter? Journal of Business Ethics 72 (2), 149– 162.

ip t

Stiglitz, J. E., 1987. Design of labor contracts: economics of incentives and risk-sharing. In: Nalbantian, H. R. (Ed.), Incentives, Cooperation, and Risk Sharing : Economic and Psychological Perspectives on Employment Contracts. Rowman & Littlefield, Totowa, N.J., pp. 47–68.

cr

Sundqvist, S.-I., 1997-2009. Owners and Powers in Sweden’s Listed Companies. SiS ¨ Agarservice AB, Stockholm, Sweden, [1997-2002 with A. Sundin, and 2003-2009 with D. Fristedt].

us

von Below, D., Thoursie, P. S., 2010. Last in, first out?: Estimating the effect of seniority rules in Sweden. Labour Economics 17 (6), 987 – 997.

te

d

M

an

Wiklund, J., 2006. Commentary: family firms and social responsibility: preliminary evidence from the S&P 500. Entrepreneurship Theory and Practice 30 (6), 803–808.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

18 Page 20 of 53

Tables Table 1 Share of family firms. Publicly listed firms Family firms Non-family firms 0.2478 0.7522

Share of Employment

0.2867

0.7133

0.2972

Observations

332,339

425,380

799

Family firms N

cr p50

p75

14.04 18,149 5,581 7.592 0.00179 0.000560 0.346 0.171

6 4,522 1,899 2 0 0 0 0

8 7,975 3,013 6 0 0 0 0

14 16,424 5,366 13 0 0 1 0

mean

p25

p50

p75

27.28 55,513 13,644 7.046 0.00447 0.00514 0.473 0.515

7 5,778 2,203 1 0 0 0 0

12 12,898 4,475 5 0 0 0 1

24 35,282 10,804 12 0 0 1 1

an

d Non-family firms N

Employment Sales Value added Age Public SOE Metro Corporate group

p25

M

332,339 317,997 317,997 332,339 332,339 332,339 316,466 332,339

te

Employment Sales Value added Age Public SOE Metro Corporate group

mean

425,380 392,653 392,653 425,380 425,380 425,380 376,719 425,380

0.7028 2,426

us

Table 2 Characteristics of family firms.

ip t

Share of firms

All firms Family firms Non-family firms 0.4386 0.5614

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: Sales and value added are measured in thousands of krona (SEK). Metro corresponds to a dummy variable for the three greater metropolitan areas around Stockholm, G¨oteborg, and Malm¨o.

19 Page 21 of 53

ip t cr

Table 3 Sensitivity of employment to industry shocks.

ln employmentist (dependent variable)

ln Sist × public ln ageist ln Sist × corp. group ln Sist × metro

Year FE Firm FE Year FE x Year FE x Year FE x Year FE x Year FE x Year FE x

-0.131 (0.187)

-0.495** (0.220) -0.0344* (0.0203) 0.0585** (0.0272)

-0.140 (0.190)

-0.413* (0.250) -0.0464 (0.0522) 0.0589* (0.0324)

yes yes yes yes yes yes yes yes

yes yes no no no no no no

yes yes yes yes yes yes yes yes

yes yes no no no no no no

yes yes yes yes yes no yes yes

yes yes no no no no no no

yes yes yes yes yes no yes yes

family firm ln age SOE public corp. group metro

Observations R-squared Number of firms

542,131 0.088 125,662

595,104 0.078 132,846

542,131 0.088 125,662

2,237 0.068 610

1,676 0.166 493

2,097 0.068 478

1,577 0.157 381

an

ln Sist × SOE

Value added shocks (S = value added) (1) (2) 0.000662 -0.0111 (0.0249) (0.0683) 0.00168 0.0413 (0.00293) (0.0318) 0.00661 0.0177 (0.0122) (0.0161) 0.0700*** 0.00866 (0.0225) (0.0250)

M

ln Sist × ln ageist

Sales shocks (S = sales) (1) (2) 0.0100 -0.0305 (0.0209) (0.0449) -0.000723 0.0431** (0.00299) (0.0217) 0.00584 0.0202 (0.0115) (0.0131) 0.0132 0.0482** (0.0317) (0.0228)

d

ln Sist × family firm

Value added shocks (S = value added) (1) (2) -0.0105*** -0.0133*** (0.00360) (0.00458) -0.000651*** -0.00912*** (0.000116) (0.00220) 0.0132*** 0.0147*** (0.00140) (0.00144) 0.0198 0.0185 (0.0235) (0.0379) 0.00952*** 0.00192 (0.00181) (0.0225) -0.149*** -0.196*** (0.0237) (0.0242) -0.000283 (0.00257) 0.00535 (0.00504)

yes yes no no no no no no

te

Variable ln Sist

Sales shocks (S = sales) (1) (2) -0.00474 -0.00688* (0.00309) (0.00393) -0.000606*** -0.00774*** (0.000108) (0.00182) 0.00711*** 0.00895*** (0.00117) (0.00119) 0.0361 0.0474 (0.0335) (0.0532) 0.00896*** -0.00307 (0.00170) (0.0220) -0.0553*** -0.110*** (0.0213) (0.0215) -0.00217 (0.00209) 0.00686 (0.00451)

Publicly listed firms

us

All firms

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

595,104 0.077 132,846

Note: Robust standard errors in parentheses. The table is split into two parts consisting of all firms (left columns) and publicly listed firms (right columns). S corresponds to the two different measures of shocks (sales and value added) used as independent variables in the left-most column. The dependent variable is ln employmentist . The estimations correspond to equation (4). *** p< 0.01, ** p< 0.05, * p< 0.1

20 Page 22 of 53

ip t cr

Table 4 First stage industry sales and value added GMM regression. ln sales at t ( S =sales ) 0.6330 (0.1154)

ln value added at t (S = value added) 0.7457 (0.1665)

ln Age Metro Corp. group Public Year FE

0.0064 -0.0169 -0.0049 0.0821 11,651

(0.0030) (0.0250) (0.0040) (0.0533) [0.000]

0.0052 -0.0184 -0.0010 0.0890 10,264

(0.0050) (0.0240) (0.0037) (0.0504) [0.000]

Hansen J-test AR(1) test AR(2) test

11.00 -5.83 0.01

[0.202] [0.000] [0.991]

6.54 -4.86 1.56

[0.162] [0.000] [0.120]

an

M

d 334,359 71,155

te

Observations Number of firms

us

Variable ln S at t − 1

334,359 71,155

Note: Dependent variable is the logarithm of total sales or value added within an industry at time t. S corresponds to sales or value added, used as independent variables in the leftmost column. Estimated by the first-differenced GMM, using robust standard errors. Sales is instrumented with lags at t−4, and value added is instrumented with 4–8 period lags. The 4–8 period lags are collapsed so that there is one instrument for each variable and lag distance. For year dummies the joint F- statistic is reported. Standard errors in parentheses. Pvalues in brackets. The estimations correspond to equation (6).

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

21 Page 23 of 53

ip t

Table 5 Sensitivity of employment to unanticipated industry shocks.

ln employmentist (dependent variable)

ǫist × family firm

ǫist × SOE

M

ǫist × public ln ageist

d

ǫist × corp. group

te

ǫist × metro

Year FE Firm FE Year FE x Year FE x Year FE x Year FE x Year FE x Year FE x

family firm ln age SOE public corp. group metro

Observations R-squared Number of firms

cr

an

ǫist × ln ageist

Value added shocks (ǫ =value added residual) (1) (2) -0.0259*** -0.0270*** (0.00475) (0.00565) -0.0132*** -0.0140*** (0.00404) (0.00426) 0.0208*** 0.0207*** (0.00241) (0.00239) 0.00168 0.00307 (0.0160) (0.0166) -0.0168 -0.0123 (0.0232) (0.0228) 0.0613*** 0.0259*** (0.00225) (0.00272) 0.00711* (0.00401) -0.00467 (0.00428)

us

Variable ǫist

Sales shocks (ǫ =sales residual) (1) (2) -0.0152*** -0.0177*** (0.00457) (0.00539) -0.0121*** -0.0131*** (0.00356) (0.00364) 0.0148*** 0.0151*** (0.00224) (0.00222) 0.0201 0.0213 (0.0362) (0.0356) -0.0318 -0.0164 (0.0283) (0.0276) 0.0618*** 0.0264*** (0.00225) (0.00272) -0.00615 (0.00382) 0.00944** (0.00424) yes yes no no no no no no

yes yes yes yes yes yes yes yes

yes yes no no no no no no

yes yes yes yes yes yes yes yes

444,266 0.062 100,620

444,266 0.070 100,620

444,266 0.062 100,620

444,266 0.071 100,620

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: Robust standard errors in parentheses. ǫ corresponds to the residuals from Table 4, and are used as independent variables in the left-most column. Dependent variable is ln employmentist . The estimations correspond to equation (5). *** p< 0.01, ** p< 0.05, * p< 0.1

22 Page 24 of 53

ip t

ln sales at t ( x =sales )

ln value added at t (x = value added)

ln employment at t (x = employment)

us

Variable

cr

Table 6 First stage firm level GMM regression on sales, value added, and employment.

0.8680 -0.1542 -0.0208 -0.0190 0.1265 3227.25 59.88

(0.0859) (0.0229) (0.0310) (0.0069) (0.1661) [0.0000] [0.0361]

0.3781 -0.0392 -0.0516 -0.0099 -0.0100 4,409 41.84

(0.0238) (0.0278) (0.0302) (0.0052) (0.1376) [0.0000] [0.4780]

0.9738 -0.0310 -0.0345 -0.0003 -0.0182 2557.27 71.45

(0.0266) (0.0031) (0.0131) (0.0022) (0.0210) [0.0000] [0.0031]

Hansen J-test AR(1) test AR(2) test AR(3) test AR(4) test AR(5) test AR(6) test AR(7) test

22.08 -10.98 6.33 -2.09 1.71 -1.03 -0.63 1.05

[0.002] [0.000] [0.000] [0.037] [0.087] [0.302] [0.528] [0.292]

0.96 -23.00 10.88 -0.71 1.54 -2.28 -0.43 2.72

[0.327] [0.000] [0.000] [0.479] [0.124] [0.023] [0.669] [0.007]

4.64 -38.91 16.80 0.77 0.05 -0.94 1.02 0.34

[0.200] [0.000] [0.000] [0.440] [0.962] [0.347] [0.309] [0.733]

Observations Number of firms

331,910 70,714

te

d

M

an

ln x at t − 1 ln Age Metro Corp. group Public Year FE Industry FE

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

328,322 69,899

371,896 75,933

Note: Dependent variable is the logarithm of firm sales, value added or employment at time t. x corresponds to sales, value added, or employment used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Sales is instrumented with lags at t − 5, and value added is instrumented with 3–4 period lags. Employment is instrumented with 6–9 period lags. The lags of value added and employment are collapsed so that there is one instrument for each variable and lag distance. For year and industry dummies the joint F- statistic is reported. Standard errors in parentheses. P-values in brackets. The estimations on sales and value added correspond to equation (7). The estimation on employment corresponds to equation (9)

23 Page 25 of 53

ip t

cr

Table 7 Sensitivity of employment to temporary and permanent firm level shocks.

Permanent shocks ∆ǫit ∆ǫit ×family firm

Observations Hansen J-test

0.00161 (0.00393) -0.0310*** (0.00675)

-7.56e-05 (0.00387) -0.0298*** (0.00663)

-0.000685 (0.00385) -0.0269*** (0.00652)

256,758 exactly identified

256,758 10.362 [0.0056]

256,758 19.162 [0.0007]

0.0482*** (0.00435) 0.00168 (0.00696)

0.0410*** (0.00437) 0.00647 (0.00759)

122,371 345.722 [0.0000]

122,371 420.257 [0.0000]

k=1

k=3

-0.0289*** (0.00460) -0.0594*** (0.00871)

-0.0250*** (0.00445) -0.0685*** (0.00847)

-0.0261*** (0.00432) -0.0654*** (0.00817)

253,923 exactly identified

253,923 46.510 [0.0000]

253,923 47.278 [0.0000]

0.0494*** (0.00395) -0.0201*** (0.00580)

0.0295*** (0.00417) -0.00604 (0.00751)

0.0284*** (0.00406) -0.000784 (0.00727)

0.0265*** (0.00401) -0.0116 (0.00713)

122,371 529.316 [0.0000]

120,866 907.464 [0.0000]

120,866 986.608 [0.0000]

120,866 1177.672 [0.0000]

an

k=2

M

Observations Hansen J-test

k=3

d

∆ǫit ×family firm

k=2

te

Temporary shocks ∆ǫit

k=1

us

∆ωit (dependent variable) Sales shocks Value added shocks (ǫ =sales residual) (ǫ =value added residual)

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: The dependent variable ∆ωit , is the first-differenced residual from the employment process in Table 6. Robust standard errors in parentheses. P-values in brackets. ǫ are the residuals from the first stage estimations on sales and value added in Table 6. For temporary shocks, instruments are defined P2 as (∆ǫi,t+1 )k for k = 1, 2, 3. For permanent shocks, instruments are defined as ( τ =−2 ∆ǫi,t+τ )k for k = 1, 2, 3. The feasible efficient GMM procedure is used for all estimations. *** p< 0.01, ** p< 0.05, * p< 0.1

24 Page 26 of 53

an

5

us

cr

Percent 10

15

ip t

20

Figures

0

20

40

60 Employment

M

5

100

Non−family firms

d

Family firms

80

te

Figure 1. Distribution of firm size, 1997-2009.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

25 Page 27 of 53

ip t

Share of family firms, % 58 − 69 55 − 58 54 − 55 52 − 54 50 − 52 49 − 50 47 − 49 43 − 47 29 − 43

te

d

M

an

us

4401 − 98806 2382 − 4401 1658 − 2382 1230 − 1658 1024 − 1230 766 − 1024 525 − 766 379 − 525 127 − 379

cr

Number of firms

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Figure 2. Geographical distribution over municipalities, 1997-2009. The enlarged areas correspond to the greater metropolitan areas of Stockholm, Gothenburg, and Malm¨o. Note: see Table B11 for details of the metropolitan areas.

26 Page 28 of 53

Appendix A

te

d

M

an

us

cr

ip t

Construction of the family firm variable To identify family-owned firms I used a methodology which has been previously described in Bjuggren et al. (2011). As a first step, the sample was restricted to incorporated firms (aktiebolag).22 There is little variation in ownership concentration within partnerships and sole-proprietorships, and legal entities such as non-profit organizations and foundations cannot by definition be family-owned in Sweden. To identify family ownership in incorporated firms, I make use of a tax reform in the early 1990s. The tax reform resulted in labor income tax for high-income earners being greater than capital income tax. In order to prevent high-income earners from exploiting this disparity, rules for closely-held firms where implemented. The rules apply only to private incorporated firms, and as of 1993 the Swedish Tax Authority classifies all private incorporated firms as closely held or not.23 A closely held firm is defined as a firm in which four or fewer owners together control more than 50 percent of the firm. To prevent owners from distributing shares to family members, the rules stipulate that family members are regarded as one owner (SFS, 1999, chapter 56). The Swedish Tax Authority registers keep track of the exact number of owners in all closely held firms. This register was matched with the firm data from Statistics Sweden and closely held firms with one or two owners where then identified as family firms to meet the 50 percent criterion. This is a conservative measure since there could be firms with three or more owners that still meet the 50 percent criterion. However, most of the closely held firms have a very concentrated ownership structure. For example, in 2006 more than 90 percent of the closely held firms had one or two owners. Finally, to identify public family-owned firms I used the standard work on ownership in public firms in Sweden, Owners and Power in Sweden’s Listed Companies by Sundqvist (1997-2009).24 Although the underlying data on closely held firms come from the Swedish Tax Authority, Statistics Sweden has made some changes in the way ownership concentration is measured. From 1997 to 2003, an owner was matched one-to-one with a single firm based on the owner’s main source of income. As of 2004, an individual is allowed to be registered as an owner of several closely held firms. This drastically increased the number of closely held firms, from 29 percent in 2003 to 49 percent in 2004. To be able to use the entire time dimension and make the data more comparable over time, I have taken the ownership information from 2004 and imputed ownership status for the period 1997-2003. If a firm is registered as a closely held firm in 2004, but not in 1997-2003, I have coded it as closely held. This hinges on the assumption that these firms have not transferred ownership classification, from a having a dispersed ownership to being closely held, in the six years before 2004. Most ownership changes are likely to take the opposite direction. A table with the number of family and non-family firms per year, using the imputed data, is presented in the online Appendix B.25

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

22

I use the term incorporated firm to describe the Swedish legal form aktiebolag, which refers to a stock corporation, sometimes also translated as limited company. 23 The Swedish Tax Authority also classifies all partnerships as closely held or not, although they do not abide by the same tax rules. 24 Table B16 in the online Appendix B shows the share of closely held firms and private family firms. 25 See Table B17.

27 Page 29 of 53

As a robustness check, all estimations are replicated using 2004-2009 as a time period.26 The results for industry shocks and temporary firm level shocks are similar to the ones in the paper. Since the time period is reduced significantly, there are not enough points in time to estimate permanent firm level shocks. The results cannot be confirmed for unanticipated shocks, since I was not able to find an adequate specification due to the reduced time period.27

(A1)

us

ǫit = ζit + ν˜it − θ˜ νi,t−1 , ζit = ζi,t−1 + u˜it .

cr

ip t

Identification of Firm Level Shocks Following Guiso et al. (2005), the error term from equation (7) can be decomposed into a random walk and an MA(1) process respectively, such that

M

an

2 2 Under the assumption that E(˜ u2it ) = σu2˜ , and E(˜ νit2 ) = σν2˜ for all t, E(˜ νis ν˜it ) = E(˜ u2is u˜2it ) = 2 2 0 for s 6= t, and E(˜ νis u˜it ) = 0 for all s and t, the autocovariance of ∆ǫit is consistent with the pattern of significance for value added in Table 6, where the first-differenced residuals become statistically insignificant after two periods. On the basis of this representation, I rewrite the equation (7) as the sum of a deterministic Dit , a permanent Pit , and a transitory component Tit :

ln Sit = Dit + Pit + Tit

(A2)

te

d

where, using L as a lag operator, (1 − ρL)−1 ǫit = Pit + Tit so that Dit = (1 − ρL)−1 (αi + θ′ Xit ). Moreover, Guiso et al. (2005) show that Pit = (1−ρL)−1 ζit , and Tit = (1−ρL)−1 ((1− θL)˜ νit − (1 − ρ)−1 ρ˜ uit ). Taking first differences, I can rewrite the error term, (1 − ρL)−1 ǫit = Pit + Tit , as ∆ǫit = (1 − ρL)uit + ∆νit

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

(A3)

where

and

uit =

1 u˜it . 1−ρ

νit = (1 − θL)˜ νit −

ρ u˜it . 1−ρ

(A4)

(A5)

As described in the main text, employment is assumed to respond to permanent and temporary shocks with the sensitivities α and β respectively. The process of employment 26

See Tables B18-B22. Because of the reduced time period I was not able to find any satisfactory lags to use as instruments in the first step GMM equations for the unanticipated industry shocks. As a default I used lagged dependent variables in period 2 and earlier as instruments. The Hansen J-tests indicate that they are potentially miss-specified. The standard errors are inflated when the residuals are used in the final estimations and no statistical significance is observed for the interaction between the family firm dummy and the shocks residual. 27

28 Page 30 of 53

takes the form ln Yit = ρ ln Yi,t−1 + (1 − ρL)Xit Φ + (1 − ρL)(ϕi + αPit + βTit + ψit )

(A6)

ωit = (1 − ρL)(ϕi + αPit + βTit + ψit ).

ip t

and the error term can be defined as (A7)

te

d

M

an

us

cr

Taking first differences and using the definition of P and T from above, together with equation A4 and A5, the residual from the process of employment can be rewritten as ∆ωit = α(1 − ρL)uit + β∆νit + (1 − ρL)∆ψit , where the fixed effects cancel out. Recall from above that the first difference of the residual from equation (7) can be written as ∆ǫit = (1 − ρL)uit + ∆νit . The two coefficients α and β are recovered by estimating two separate IV regressions of ∆ωit on ∆ǫit . As mentioned in thePmain text, Guiso et al. (2005) show that α can be estimated based on the instruments ( 2τ =−2 ∆ǫi,t+τ )k , and β can be estimated based on the instruments (∆ǫi,t+1 )k , for any k ≥ 1.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

29 Page 31 of 53

Online Appendix B Data construction The data used are register data from Statistics Sweden (SCB) on all firms with at least five employees, and covers the period from 1997 to 2009. For firms with less than 5 em-

ip t

ployees there is little variation in ownership concentration. The data are obtained from various register sources. Number of employees, legal form, corporate group information,

cr

firm age, industries, and geographical location is obtained from the RAMS-register (Labor Statistics Based on Administrative Sources), and F¨oretagsdatabasen (Statistics Sweden’s Business Register). Data on sales and value added are obtained from F¨oretagens Ekonomi

us

(the Structural Business Statistics), and deflated using the fixed consumer price index (CPI) from Statistics Sweden. Information on closely held firms is obtained from the Swedish Tax

an

Authority, and information on family-owners in public firms is obtained from the annual compilation ”Owners and Power in Sweden’s Listed Companies” by Sven-Ivan Sundqvist. Number of employees is defined according to the number of employees in November earning

te

d

M

a salary that exceeds a certain threshold.28

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

28

To determine the threshold, individuals are divided into 25 categories depending on variables such as age, gender, and retirement pension. As an example, in 2005, for a male of age 25-54, the threshold is an annual salary of 50,036 SEK (Statistics Sweden, 2009). This is equivalent of about 6,255 USD, using the exchange rate on November 1, 2005.

30 Page 32 of 53

Tables Table B1 Sensitivity of employment to industry shocks, including firms with more than 500 employees. ln employmentist (dependent variable)

ln xist ln xist × corp. group ln xist × metro

Year FE Firm FE Year FE x Year FE x Year FE x Year FE x Year FE x Year FE x

family firm ln age SOE public corp. group metro

Observations R-squared Number of firms

yes yes yes yes yes yes yes yes

yes yes yes yes yes no yes yes

yes yes yes yes yes no yes yes

444,781 0.071 100,694

444,212 0.071 100,625

542,655 0.088 125,727

cr

yes yes yes yes yes yes yes yes

us

ln xist × public

Value added shocks residual (x = value added residual) -0.0266*** (0.00587) -0.0210*** (0.00424) 0.0250*** (0.00244) -0.0169 (0.0196) -0.0181 (0.0116) 0.0257*** (0.00273) 0.00474 (0.00424) 0.00474 (0.00424)

an

ln xist × SOE

Sales shocks residual (x = sales residual) -0.0158*** (0.00549) -0.0159*** (0.00345) 0.0164*** (0.00218) 0.00335 (0.0382) -0.000672 (0.0235) 0.0263*** (0.00272) 0.00987** (0.00452) 0.00987** (0.00452)

M

ln xist × ln ageit

Value added shocks (x = value added) -0.0121*** (0.00462) -0.0105*** (0.00221) 0.0163*** (0.00145) -0.000304 (0.0358) -0.0176 (0.0166) -0.226*** (0.0248) 0.00262 (0.00500) 0.00262 (0.00500)

d

ln xist × family firm

Sales shocks (x = sales) -0.00546 (0.00402) -0.00884*** (0.00182) 0.00976*** (0.00119) 0.0370 (0.0567) 0.00542 (0.0201) -0.127*** (0.0219) 0.00672 (0.00461) 0.00672 (0.00461)

te

Variable ln xist

Unanticipated industry shocks

ip t

Industry shocks

542,240 0.089 125,669

Note: Robust standard errors in parentheses. The table is split into two parts according to industry shocks and unanticipated shocks. Industry shocks correspond to equation (4), and unanticipated industry shocks correspond to equation (5). x corresponds to the two different measures of shocks (sales and value added) used as independent variables in the left-most column. The dependent variable is ln employmentist . Residuals used to estimate unanticipated shocks are from the first stage equations in Table B2. *** p< 0.01, ** p< 0.05, * p< 0.1

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

31 Page 33 of 53

ip t cr

us

Table B2 First stage industry sales and value added GMM regression, including firms with more than 500 employees. ln sales at t ( S =sales ) 0.5707 (0.1163)

ln value added at t (S = value added) 0.6978 (0.1247)

ln Age Metro Public Corp. group Year FE

0.0074 -0.0287 0.0687 -0.0078 12,229

(0.0030) (0.0243) (0.0503) (0.0039 ) [0.000]

0.0067 -0.0335 0.0352 -0.0030 10,030

(0.0042) (0.0220) (0.0562) (0.0034) [0.000]

Hansen J-test AR(1) test AR(2) test

28.20 -5.45 1.02

[0.000] [0.000] [0.308]

17.09 -6.24 2.75

[0.002] [0.000] [0.006]

te

d

M

an

Variable ln S at t − 1

Observations Number of firms

334,793 71,215

334,210 71,164

Note: Dependent variable is the logarithm of total sales or value added within an industry at time t. S corresponds to sales or value added, used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Instruments are defined as in Table 4. For year dummies the joint F- statistic is reported. Standard errors in parentheses. P-values in brackets. The estimations correspond to equation (6).

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

32 Page 34 of 53

ip t

ln sales at t ( x =sales )

ln value added at t (x = value added)

ln employment at t (x = employment)

us

Variable

cr

Table B3 First stage firm level GMM regression on sales, value added, and employment, including firms with more than 500 employees.

0.8759 -0.1562 -0.0218 -0.0194 0.1237 3,226 59.59

(0.0854) (0.0228) (0.0310) (0.0070) (0.1651) [0.0000] [0.0382]

0.3829 -0.0405 -0.0525 -0.0101 -0.0088 4,402 41.91

(0.0237) (0.0071) (0.0302) (0.0052) (0.1362) [0.0000] [0.4750]

0.9759 -0.0311 -0.0350 -0.0004 -0.0187 2,579 71.55

(0.0263) (0.0031) (0.0131) (0.0022) (0.0209) [0.0000] [0.0030]

Hansen J-test AR(1) test AR(2) test AR(3) test AR(4) test AR(5) test AR(6) test AR(7) test

22.61 -11.07 6.39 -2.14 1.63 -0.99 -0.67 1.00

[0.002] [0.000] [0.000] [0.033] [0.103] [0.322] [0.505] [0.315]

0.94 -23.05 10.96 -0.67 1.36 -2.02 -0.61 2.72

[0.332] [0.000] [0.000] [0.500] [0.172] [0.043] [0.541] [0.007]

4.84 -39.44 16.85 0.77 0.06 -0.96 1.03 0.32

[0.184] [0.000] [0.000] [0.439] [0.954] [0.335] [0.304] [ 0.751]

Observations Number of firms

M

d

te

332,342 70,774

an

ln x at t − 1 ln Age Metro Corp. group Public Year FE Industry FE

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

328,749 69,957

372,363 75,995

Note: Dependent variable is the logarithm of firm sales, value added or employment at time t. x corresponds to sales, value added, or employment used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Instruments are defined as in Table 6. For year and industry dummies the joint F- statistic is reported. Standard errors in parentheses. P-values in brackets. The estimations on sales and value added correspond to equation (7). The estimation on employment corresponds to equation (9)

33 Page 35 of 53

ip t

cr

Table B4 Sensitivity of employment to temporary and permanent firm level shocks, including firms with more than 500 employees.

Permanent shocks ∆ǫit ∆ǫit ×family firm

Observations Hansen J-test

0.00193 (0.00388) -0.0309*** (0.00670)

0.000185 (0.00382) -0.0296*** (0.00658)

-0.000497 (0.00379) -0.0265*** (0.00645)

257,126 exactly identified

257,126 10.197 [0.0061]

k=1

k=3

-0.0285*** (0.00455) -0.0586*** (0.00866)

-0.0247*** (0.00441) -0.0670*** (0.00843)

-0.0258*** (0.00428) -0.0647*** (0.00812)

257,126 19.059 [0.0008]

254,287 exactly identified

254,287 45.516 [0.0000]

254,287 45.656 [0.0000]

an

k=2

M

Observations Hansen J-test

k=3

d

∆ǫit ×family firm

k=2

te

Temporary shocks ∆ǫit

k=1

us

∆ωit (dependent variable) Sales shocks Value added shocks (ǫ =sales residual) (ǫ =value added residual)

0.0483*** (0.00435) 0.00148 (0.00694)

0.0411*** (0.00443) 0.00623 (0.00763)

0.0513*** (0.00400) -0.0222*** (0.00584)

0.0297*** (0.00415) -0.00711 (0.00746)

0.0289*** (0.00404) -0.00255 (0.00723)

0.0273*** (0.00400) -0.0126* (0.00711)

122,602 343.561 [0.0000]

122,602 417.664 [0.0000]

122,602 520.352 [0.0000]

121,094 888.967 [0.0000]

121,094 961.931 [0.0000]

121,094 1,168.814 [0.0000]

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: The dependent variable ∆ωit , is the first-differenced residual from the employment process in Table B3. Robust standard errors in parentheses. P-values in brackets. ǫ are the residuals from the first stage estimations on sales and value added in Table B3. For temporary shocks, instruments are defined P2 as (∆ǫi,t+1 )k for k = 1, 2, 3. For permanent shocks, instruments are defined as ( τ =−2 ∆ǫi,t+τ )k for k = 1, 2, 3. The feasible efficient GMM procedure is used for all estimations. *** p< 0.01, ** p< 0.05, * p< 0.1

34 Page 36 of 53

Table B5 Firms within each industry.

manufacture of food products and beverages manufacture of textiles manufacture of wearing apparel tanning and dressing of leather manufacture of wood and of products of wood and cork manufacture of paper and paper products publishing, printing and reproduction of recorded media manufacture of chemicals and chemical products manufacture of rubber and plastics products manufacture of other non-metallic mineral products manufacture of basic metals manufacture of fabricated metal products manufacture of machinery and equipment n.e.c.* manufacture of office, accounting and computing machinery manufacture of electrical machinery and apparatus n.e.c. manufacture of radio, television and communication equipment manufacture of medical, precision and optical instruments, etc. manufacture of motor vehicles, trailers and semi-trailers manufacture of other transport equipment manufacture of furniture; manufacturing n.e.c. recycling construction sale, maintenance and repair of motor vehicles and motorcycles wholesale trade and commission trade retail trade, except of motor vehicles and motorcycles hotels and restaurants land transport; transport via pipelines water transport air transport supporting and auxiliary transport activities post and telecommunications financial intermediation, except insurance and pension funding activities auxiliary to financial intermediation real estate activities renting of machinery and equipment computer and related activities research and development other business activities education health and social work sewage and refuse disposal, sanitation and similar activities recreational, cultural and sporting activities other service activities

56 52 57 52 56 77 67 76 62 65 75 51 61 61 60 63 68 64 61 56 58 43 45 65 46 59 43 75 75 72 77 89 79 62 57 77 84 63 55 56 61 68 44

an

M

d

te

Total

Share of family firms

cr

15 17 18 19 20 21 22 24 25 26 27 28 29 30 31 32 33 34 35 36 37 45 50 51 52 55 60 61 62 63 64 65 67 70 71 72 73 74 80 85 90 92 93

us

Industry

ip t

Share of non-family firms

NACE

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

56

Total number of firms

44 48 43 48 44 23 33 24 38 35 25 49 39 39 40 37 32 36 39 44 42 57 55 35 54 41 57 25 25 28 23 11 21 38 43 23 16 37 45 44 39 32 56

12,081 2,948 923 516 12,658 2,392 16,660 3,536 7,585 3,574 2,271 36,063 18,963 958 5,315 2,099 5,143 3,669 2,725 7,770 1,031 98,285 36,433 88,610 88,835 46810 42,664 1,582 485 11,280 1,717 1,959 3,894 13,945 4,989 25,737 2,731 90,565 10,019 18,802 1,794 10,877 6,826

44

757,719

35 Page 37 of 53

Table B6 Mean, standard deviation, and coefficient of variation for sales and value added within each industry. ln Sales

an

M

d

Note: CV stands for the coefficient of variation, CV =

CV 0.009 0.009 0.007 0.014 0.007 0.010 0.004 0.011 0.006 0.013 0.010 0.008 0.007 0.010 0.005 0.007 0.013 0.011 0.014 0.006 0.022 0.015 0.007 0.005 0.007 0.011 0.010 0.008 0.024 0.015 0.020 0.030 0.065 0.005 0.013 0.010 0.013 0.010 0.022 0.021 0.010 0.010 0.021 0.062

Mean 16.19 14.66 13.31 12.73 16.23 15.75 16.51 16.21 16.05 15.53 15.52 17.04 17.02 14.16 15.58 14.78 15.90 15.67 14.75 15.77 13.68 17.66 16.82 18.15 17.21 16.48 16.89 15.08 13.71 16.21 14.89 13.90 13.70 16.58 15.22 17.16 14.68 18.17 15.48 16.36 14.30 15.50 14.35 17.06

Std.Dev. 0.106 0.148 0.147 0.157 0.137 0.192 0.0840 0.263 0.0776 0.193 0.110 0.104 0.0798 0.194 0.140 0.154 0.254 0.104 0.160 0.0824 0.373 0.266 0.132 0.123 0.148 0.181 0.201 0.343 0.295 0.319 0.504 0.929 0.865 0.0868 0.219 0.219 0.273 0.248 0.386 0.364 0.121 0.257 0.295 0.954

ip t

Std.Dev. 0.158 0.143 0.103 0.194 0.120 0.172 0.0616 0.190 0.104 0.210 0.167 0.147 0.127 0.161 0.0892 0.121 0.212 0.187 0.224 0.100 0.331 0.286 0.130 0.107 0.125 0.183 0.170 0.136 0.366 0.275 0.317 0.441 0.957 0.0918 0.203 0.185 0.213 0.196 0.346 0.358 0.158 0.179 0.318 1.135

ln Value added

cr

Mean 17.66 15.74 14.64 13.90 17.63 17.07 17.55 17.48 17.15 16.67 17.06 18.02 18.12 15.36 16.75 16.17 16.84 16.88 15.99 16.88 15.22 18.70 19.00 20.10 18.92 17.37 17.77 16.67 15.44 18.18 16.21 14.62 14.64 17.32 16.10 17.91 15.80 19.10 16.08 16.75 15.16 17.16 15.01 18.31

us

Industry manufacture of food products and beverages manufacture of textiles manufacture of wearing apparel tanning and dressing of leather manufacture of wood and of products of wood and cork manufacture of paper and paper products publishing, printing and reproduction of recorded media manufacture of chemicals and chemical products manufacture of rubber and plastics products manufacture of other non-metallic mineral products manufacture of basic metals manufacture of fabricated metal products manufacture of machinery and equipment n.e.c.* manufacture of office, accounting and computing machinery manufacture of electrical machinery and apparatus n.e.c. manufacture of radio, television and communication equipment manufacture of medical, precision and optical instruments, etc. manufacture of motor vehicles, trailers and semi-trailers manufacture of other transport equipment manufacture of furniture; manufacturing n.e.c. recycling construction sale, maintenance and repair of motor vehicles and motorcycles wholesale trade and commission trade retail trade, except of motor vehicles and motorcycles hotels and restaurants land transport; transport via pipelines water transport air transport supporting and auxiliary transport activities post and telecommunications financial intermediation, except insurance and pension funding activities auxiliary to financial intermediation real estate activities renting of machinery and equipment computer and related activities research and development other business activities education health and social work sewage and refuse disposal, sanitation and similar activities recreational, cultural and sporting activities other service activities

te

NACE 15 17 18 19 20 21 22 24 25 26 27 28 29 30 31 32 33 34 35 36 37 45 50 51 52 55 60 61 62 63 64 65 67 70 71 72 73 74 80 85 90 92 93 Total

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

CV 0.007 0.010 0.011 0.012 0.008 0.012 0.005 0.016 0.005 0.012 0.007 0.006 0.005 0.014 0.009 0.010 0.016 0.007 0.011 0.005 0.027 0.015 0.008 0.007 0.009 0.011 0.012 0.023 0.022 0.020 0.034 0.067 0.063 0.005 0.014 0.013 0.019 0.014 0.025 0.022 0.008 0.017 0.021 0.056

Std.Dev. Mean .

36 Page 38 of 53

ip t

Table B7 Sensitivity of employment to industry shocks in private firms.

ln employmentist (dependent variable)

ln ageist ln xist × corp. group ln xist × metro

Year FE Firm FE Year FE x Year FE x Year FE x Year FE x Year FE x Year FE x

family firm ln age SOE public corp. group metro

yes yes yes yes yes yes yes yes

Observations R-squared Number of firms

yes yes yes yes yes yes yes yes

yes yes yes yes yes yes yes yes

yes yes yes yes yes yes yes yes

540,654 0.087 125,388

540,654 0.088 125,388

442,884 0.070 100,355

442,884 0.070 100,355

us

Value added shocks residual (x = value added residual) -0.0269*** (0.00576) -0.0144*** (0.00436) 0.0208*** (0.00243) -0.00502 (0.0166) 0.0258*** (0.00273) -0.00509 (0.00435) 0.00784* (0.00407)

an

ln xist × SOE

Sales shocks residual (x = sales residual) -0.0167*** (0.00535) -0.0137*** (0.00389) 0.0144*** (0.00224) 0.00263 (0.0289) 0.0261*** (0.00273) -0.00709* (0.00390) 0.00882** (0.00383)

M

ln xist × ln ageist

Value added shocks (x = value added) -0.0128*** (0.00467) -0.00953*** (0.00228) 0.0150*** (0.00147) 0.000329 (0.0388) -0.199*** (0.0247) -0.000415 (0.00261) 0.00576 (0.00514)

d

ln xist × family firm

Sales shocks (x = sales) -0.00597 (0.00398) -0.00802*** (0.00187) 0.00887*** (0.00120) 0.0302 (0.0610) -0.108*** (0.0217) -0.00236 (0.00210) 0.00718 (0.00456)

te

Variable ln xist

Unanticipated industry shocks

cr

Industry shocks

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: Robust standard errors in parentheses. The table is split into two parts according to industry shocks and unanticipated shocks. Industry shocks correspond to equation (4), and unanticipated industry shocks correspond to equation (5). x corresponds to the two different measures of shocks (sales and value added) used as independent variables in the left-most column. The dependent variable is ln employmentist . Residuals used to estimate unanticipated shocks are from the first stage equations in Table B8. *** p< 0.01, ** p< 0.05, * p< 0.1

37 Page 39 of 53

ip t cr

us

Table B8 First stage industry sales and value added GMM regression for private firms. ln sales at t ( S =sales ) 0.7358 (0.1234)

ln value added at t (S = value added) 0.7540 (0.1588)

ln Age Metro Corp. group Year FE

0.0047 -0.0175 -0.0048 10,928

(0.0031) (0.0259) (0.0041) [0.000]

0.0059 -0.0192 -0.0004 9,815

(0.0048) (0.0235) (0.0036) [0.000]

Hansen J-test AR(1) test AR(2) test

9.88 -6.21 -0.08

[0.274] [0.000] [0.939]

7.89 -5.11 1.67

[0.096] [0.000] [0.094]

te

d

M

an

Variable ln S at t − 1

Observations Number of firms

333,276 70,943

333,276 70,943

Note: Dependent variable is the logarithm of total sales or value added within an industry at time t. S corresponds to sales or value added, used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Instruments are defined as in Table 4. For year dummies the joint F- statistic is reported. Standard errors in parentheses. P-values in brackets. The estimations correspond to equation (6).

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

38 Page 40 of 53

ip t

ln sales at t ( x =sales )

ln value added at t (x = value added)

ln employment at t (x = employment)

us

Variable

cr

Table B9 First stage firm level GMM regression on sales, value added, and employment for private firms.

0.8555 -0.1516 -0.0214 -0.0180 3,302 61.89

(0.0832) (0.0222) (0.0310) (0.0068) [0.0000] [0.0245]

0.3767 -0.0391 -0.0493 -0.0096 4,442 39.96

(0.0236) (0.0070) (0.0301) (0.0052) [0.0000] [0.5608]

0.9725 -0.0307 -0.0356 -0.0003 2,560 69.24

(0.0266) (0.0031) (0.0131) (0.0022) [0.0000] [0.0051]

Hansen J-test AR(1) test AR(2) test AR(3) test AR(4) test AR(5) test AR(6) test AR(7) test

22.00 -11.18 6.66 -1.37 1.32 -0.87 -0.70 1.13

[0.003] [0.000] [0.000] [0.170] [0.187] [0.384] [0.482] [0.257]

1.14 -23.10 11.05 -0.76 1.56 -2.45 -0.29 2.72

[0.285] [0.000] [0.000] [0.447] [0.120] [0.014] [0.770] [0.007]

4.08 -38.91 16.90 0.56 0.17 -0.99 1.08 0.32

[0.253] [0.000] [0.000] [0.577] [0.864] [0.321] [0.281] [ 0.747]

Observations Number of firms

M

d

te

331,013 70,526

an

ln x at t − 1 ln Age Metro Corp. group Year FE Industry FE

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

327,786 69,776

370,551 75,689

Note: Dependent variable is the logarithm of firm sales, value added or employment at time t. x corresponds to sales, value added, or employment used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Instruments are defined as in Table 6. For year and industry dummies the joint F- statistic is reported. Standard errors in parentheses. P-values in brackets. The estimations on sales and value added correspond to equation (7). The estimation on employment corresponds to equation (9)

39 Page 41 of 53

ip t

cr

Table B10 Sensitivity of employment to temporary and permanent firm level shocks in private firms.

Permanent shocks ∆ǫit ∆ǫit ×family firm

Observations Hansen J-test

0.000650 (0.00410) -0.0303*** (0.00700)

-0.00114 (0.00403) -0.0291*** (0.00691)

-0.00132 (0.00402) -0.0297*** (0.00684)

256,065 exactly identified

256,065 14.256 [0.0008]

256,065 23.737 [0.0001]

0.0512*** (0.00450) -0.00946 (0.00715)

0.0446*** (0.00446) 0.000664 (0.00790)

122,060 387.810 [0.0000]

122,060 446.211 [0.0000]

k=1

k=3

-0.0298*** (0.00466) -0.0588*** (0.00875)

-0.0260*** (0.00451) -0.0678*** (0.00851)

-0.0271*** (0.00437) -0.0648*** (0.00821)

253,524 exactly identified

253,524 45.891 [0.0000]

253,524 46.561 [0.0000]

0.0501*** (0.00405) -0.0102 (0.00668)

0.0294*** (0.00420) -0.00530 (0.00752)

0.0282*** (0.00408) 0.000179 (0.00727)

0.0261*** (0.00403) -0.0105 (0.00713)

122,060 533.468 [0.0000]

120,695 907.882 [0.0000]

120,695 987.996 [0.0000]

120,695 1,181.958 [0.0000]

an

k=2

M

Observations Hansen J-test

k=3

d

∆ǫit ×family firm

k=2

te

Temporary shocks ∆ǫit

k=1

us

∆ωit (dependent variable) Sales shocks Value added shocks (ǫ =sales residual) (ǫ =value added residual)

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: The dependent variable ∆ωit , is the first-differenced residual from the employment process in Table B9. Robust standard errors in parentheses. P-values in brackets. ǫ are the residuals from the first stage estimations on sales and value added in Table B9. For temporary shocks, instruments are defined P2 as (∆ǫi,t+1 )k for k = 1, 2, 3. For permanent shocks, instruments are defined as ( τ =−2 ∆ǫi,t+τ )k for k = 1, 2, 3. The feasible efficient GMM procedure is used for all estimations. *** p< 0.01, ** p< 0.05, * p< 0.1

40 Page 42 of 53

ip t

cr

Table B11 The three greater metropolitan areas, divided into municipalities. G¨oteborg

Malm¨o

Botkyrka Danderyd Eker¨ o Haninge Huddinge J¨arf¨ alla Liding¨ o Nacka Norrt¨ alje Nykvarn Nyn¨ ashamn Salem Sigtuna Sollentuna Solna Stockholm Sundbyberg S¨ odert¨ alje Tyres¨ o T¨ aby Upplands Upplands-Bro Vallentuna Vaxholm V¨ armd¨ o ¨ Oster˚ aker

Ale Alings˚ as G¨oteborg H¨ arryda Kungsbacka Kung¨alv Lerum Lilla Edet M¨ olndal Partille Stenungsund Tj¨orn ¨ Ocker¨ o

Burl¨ ov Esl¨ov H¨ oo¨r K¨avlinge Lomma Lund Malm¨o Skurup Staffanstorp Svedala Trelleborg Vellinge

te

d

M

an

us

Stockholm

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: The three greater metropolitan areas are defined by Statistics Sweden. They are also referred to as: Stor-Stockholm, Stor- G¨ oteborg and Stor-Malm¨ o (Stor being the Swedish word for great).

41 Page 43 of 53

ip t

Table B12 Sensitivity of employment to industry shocks – alternate model specification.

cr

ln employmentist (dependent variable) All firms

ln Sist × public ln ageist

Year FE Firm FE Year FE x Year FE x Year FE x Year FE x Year FE x Year FE x

family firm ln age SOE public corp. group metro

Observations R-squared Number of firms

Value added shocks (S = value added) 0.00129 (0.0287) 0.00907 (0.0225) 0.00542 (0.0129) 0.0259 (0.0179)

-0.221 (0.191)

-0.205 (0.212)

yes yes yes yes yes yes no no

yes yes yes yes yes yes no no

yes yes yes yes yes yes no no

yes yes yes yes yes yes no no

595,104 0.083 132,846

595,104 0.084 132,846

2,237 0.098 610

2,097 0.093 478

us

Sales shocks (S = sales) 0.000911 (0.0230) 0.0230 (0.0176) 0.00542 (0.0109) 0.00599 (0.0179)

an

ln Sist × SOE

Value added shocks (S = value added) -0.0104*** (0.00371) -0.00946*** (0.00215) 0.0140*** (0.00140) 0.0210 (0.0290) 0.0171 (0.0227) -0.186*** (0.0236)

M

ln Sist × ln ageist

Sales shocks (S = sales) -0.00437 (0.00318) -0.00709*** (0.00179) 0.00776*** (0.00117) 0.0414 (0.0397) 0.0179 (0.0221) -0.0906*** (0.0211)

d

ln Sist × family firm

Publicly listed firms

te

Variable ln Sist

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: Robust standard errors in parentheses. The table is split into two parts consisting of all firms (left columns) and publicly listed firms (right columns). S corresponds to the two different measures of shocks (sales and value added) used as independent variables in the left-most column. The dependent variable is ln employmentist . The estimations correspond to equation (4). *** p< 0.01, ** p< 0.05, * p< 0.1

42 Page 44 of 53

ip t cr

us

Table B13 First stage industry sales and value added GMM regression for publicly listed firms. ln sales at t ( S =sales ) 0.0215 (0.4590)

ln Age Metro Corp. group Year FE

-0.1057 0.0009 0.1591 58.01

(0.1620) (0.0895) (0.1483) [0.000]

-0.0702 -0.0632 -0.0590 83.27

(0.1463) (0.2510) (0.1820) [0.000]

Hansen J-test AR(1) test AR(2) test

4.95 -0.33 -0.88

[0.176] [0.740] [0.381]

3.94 -1.16 -1.01

[0.268] [0.246] [0.312]

M

d 921 259

te

Observations Number of firms

ln value added at t (S = value added) 0.3716 (0.4314)

an

Variable ln S at t − 1

828 248

Note: Dependent variable is the logarithm of total sales or value added within an industry at time t. S corresponds to sales or value added, used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Sales is instrumented with 4–7 period lags, and value added is instrumented with 5–8 period lags, The lags are collapsed so that there is one instrument for each variable and lag distance. For year dummies the joint Fstatistic is reported. Standard errors in parentheses. P-values in brackets. The estimations correspond to equation (6).

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

43 Page 45 of 53

Variable

Estimate 2.4030 -0.5410 -0.0765 -0.0568 0.0723 720.09 38.19

(0.3540) (0.0927) (0.0580) (0.0148) (0.3111) [0.0000] [0.6390]

Hansen J-test AR(1) test AR(2) test AR(3) test AR(4) test AR(5) test AR(6) test AR(7) test

5.31 -6.91 4.93 -1.80 1.95 -0.36 -0.06 1.02

[0.380] [0.000] [0.000] [0.073] [0.051] [0.720] [0.951] [0.305]

Observations Number of firms

331,910 70,714

ip t

an

M

d

te

us

ln Sales at t − 1 ln Age Metro Corp. group Public Year FE Industry FE

cr

Table B14 Firm level sales GMM regression using lags 7 and earlier.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: Dependent variable is the logarithm of sales at time t. Estimated by the first-differenced GMM, using robust standard errors. For year and industry dummies the joint F- statistic is reported. Standard errors in parentheses. P-values in brackets. Instrumented with sales at t − 7 and earlier. The lags are collapsed so that there are one instrument for each variable and lag distance. The estimation corresponds to equation (7).

44 Page 46 of 53

Table B15 Sensitivity of employment to firm level sales shocks using alternate residuals. ∆ωit (dependent variable) Temporary

Observations Hansen J-test

k=1

k=2

k=3

0.0220*** (0.00254) -0.0132*** (0.00406)

0.0197*** (0.00243) -0.0103*** (0.00376)

0.0172*** (0.00231) -0.00929*** (0.00343)

0.0270*** (0.00275) 0.00602 (0.00421)

0.0287*** (0.00276) -0.00318 (0.00429)

0.0276*** (0.00231) -0.00856*** (0.00323)

256,758 exactly identified

256,758 13.639 [0.0011]

256,758 38.937 [0.0000]

122,371 589.219 [0.0000]

ip t

k=3

cr

∆ǫit ×family firm

k=2

122,371 694.608 [0.0000]

us

∆ǫit

Permanent

k=1

122,371 826.443 [0.0000]

an

Note: The dependent variable ∆ωit , is the first-differenced residual from the employment process in Table 6. Robust standard errors inP parentheses. P-values in brackets. Instruments for the permanent 2 shock estimations are defined as ( τ =−2 ∆ǫi,t+τ )k for k = 1, 2, 3.. Instruments for the temporary shock estimations are defined as (∆ǫi,t+1 )k for k = 1, 2, 3. ǫ is the residual from the estimation in Table B14. *** p< 0.01, ** p< 0.05, * p< 0.1

M

Table B16 Share of closely held firms and private family firms.

Observations

347,531

Private firms

non-closely held firms 0.5413

family 0.4393

non-family 0.5607

410,188

331,745

423,478

te

d

Share of firms

closely held firms 0.4587

Table B17 Number of firms, year by year.

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Year

Nonfamily

Family

Total

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

31,284 32,189 32,793 33,646 32,983 30,851 29,634 31,026 31,755 33,186 34,727 35,832 35,474

19,994 20,932 21,764 22,490 23,195 25,636 26,746 26,993 27,795 28,228 29,421 29,873 29,272

51,278 53,121 54,557 56,136 56,178 56,487 56,380 58,019 59,550 61,414 64,148 65,705 64,746

Total

425,380

332,339

757,719

45 Page 47 of 53

ip t

Table B18 Sensitivity of employment to industry shocks 2004–2009. ln employmentist (dependent variable)

ln Sist × SOE ln Sist × public ln ageist ln Sist × corp. group

Year FE Firm FE Year FE x Year FE x Year FE x Year FE x Year FE x Year FE x

0.363 (0.378) -0.108* (0.0584) -0.00152 (0.0274)

0.354 (0.335) -0.136** (0.0606) 0.00268 (0.0192)

yes yes yes yes yes yes yes yes

yes yes yes yes yes yes yes yes

yes yes yes yes yes no yes yes

yes yes yes yes yes no yes yes

family firm ln age SOE public corp. group metro

Observations R-squared Number of firms

273,522 0.058 83,884

273,522 0.059 83,884

841 0.150 295

796 0.159 291

us

Value added shocks (S = value added) 0.182** (0.0733) 0.0234 (0.0306) -0.0306 (0.0206) 0.0549*** (0.0181)

te

ln Sist × metro

Sales shocks (S = sales) 0.148* (0.0784) 0.0528* (0.0317) -0.0276 (0.0221) 0.0896*** (0.0232)

an

ln Sist × ln ageit

Value added shocks (S = value added) -0.0207*** (0.00543) -0.00532*** (0.00189) 0.0181*** (0.00166) 0.0167 (0.0331) 0.0266* (0.0140) -0.309*** (0.0285) -0.00464 (0.00307) 0.00695 (0.00533)

M

ln Sist × family firm

Sales shocks (S = sales) -0.00431 (0.00459) -0.00430*** (0.00158) 0.00800*** (0.00137) 0.0321 (0.0350) 0.0235 (0.0156) -0.148*** (0.0254) -0.00419 (0.00259) 0.00582 (0.00473)

d

Variable ln Sist

Public firms

cr

All firms

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: Robust standard errors in parentheses. The table is split into two parts consisting of all firms (left columns) and publicly listed firms (right columns). x corresponds to the two different measures of shocks (sales and value added) used as independent variables in the left-most column. The dependent variable is ln employmentist . The estimations correspond to equation (4). *** p< 0.01, ** p< 0.05, * p< 0.1

46 Page 48 of 53

ip t cr

us

Table B19 First stage industry sales and value added GMM regression, 2004-2009. ln sales at t ( S =sales ) 0.9511 (0.0570)

ln value added at t (S = value added) 1.0568 (0.1315)

ln Age Metro Public Corp. group Year FE

0.0075 0.0001 0.0733 -0.0077 4,648

(0.0048) (0.0358) (0.0497) (0.0087) [0.000]

0.0062 -0.0292 0.0813 -0.0028 4,355

(0.0054) (0.0385) (0.0511) (0.0079) [0.000]

Hansen J-test AR(1) test AR(2) test

57.09 -16.97 0.31

[0.000] [0.000] [0.755]

0.55 -12.76 0.80

[0.907] [0.000] [0.425]

te

d

M

an

Variable ln S at t − 1

Observations Number of firms

135,865 48,461

135,865 48,461

Note: Dependent variable is the logarithm of total sales or value added within an industry at time t. S corresponds to sales or value added, used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Both sales and value added are instrumented with lags at t − 2. For year dummies the joint F- statistic is reported. Standard errors in parentheses. Pvalues in brackets. The estimations correspond to equation (6).

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

47 Page 49 of 53

ip t

Table B20 Sensitivity of employment to unanticipated industry shocks 2004– 2009. ln employmentist (dependent variable)

ǫist × family firm

ǫist × SOE

ln ageist

yes yes yes yes yes yes yes yes

204,606 0.041 67,045

204,606 0.041 67,045

d

ǫist × corp. group

te

ǫist × metro

family firm ln age SOE public corp. group metro

Observations R-squared Number of firms

cr

yes yes yes yes yes yes yes yes

M

ǫist × public

Year FE Firm FE Year FE x Year FE x Year FE x Year FE x Year FE x Year FE x

Value added shocks residual (ǫ = value added) -0.00372 (0.00661) 0.00284 (0.00441) 0.00314 (0.0023) -0.0197 (0.0376) -0.00454 (0.0186) -0.0243*** (0.00524) 0.0100** (0.00458) -0.00870** (0.00436)

an

ǫist × ln ageist

Sales residual (ǫ = sales) -0.00393 (0.00667) -0.00129 (0.00430) 0.00413* (0.00226) -0.0282 (0.0439) -0.00504 (0.0194) -0.0243*** (0.00524) 0.00646 (0.00441) -0.00552 (0.00423)

us

Variable ǫist

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: Robust standard errors in parentheses. ǫ corresponds to the residuals from Table B19, and are used as independent variables in the left-most column. Dependent variable is ln employmentist . The estimations correspond to equation (5). *** p< 0.01, ** p< 0.05, * p< 0.1

48 Page 50 of 53

ip t

Variable

ln sales at t ( x =sales )

us

cr

Table B21 First stage firm level GMM regression on sales, value added, and employment, 20042009. ln value added at t (x = value added)

ln employment at t (x = employment)

0.7115 -0.1040 -0.0438 -0.0211 0.2432 1,741 444.57

(0.0569) (0.0148) (0.0468) (0.0129) (0.2587) [0.0000] [0.0000]

0.5413 -0.0691 -0.0947 -0.0140 -0.3002 2,720 420.85

(0.0485) (0.0139) (0.0588) (0.0108) (0.2852) [0.0000] [0.0000]

1.1942 -0.0466 -0.0221 -0.0037 -0.0172 1,623 2,976.60

(0.0722) (0.0069) (0.0231) (0.0047) (0.0366) [0.0000] [0.0000]

Hansen J-test AR(1) test AR(2) test AR(3) test

20.59 -11.34 2.13 -0.09

[0.001] [0.000] [0.033] [0.927]

8.92 -13.61 5.03 0.41

[0.112] [0.000] [0.000] [0.680]

8.52 -17.58 7.56 1.15

[0.014] [0.000] [0.000] [0.251]

Observations Number of firms

134,998 48,186

te

d

M

an

ln x at t − 1 ln Age Metro Corp. group Public Year FE Industry FE

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

133,469 47,643

147,510 50,853

Note: Dependent variable is the logarithm of firm sales, value added or employment at time t. x corresponds to sales, value added, or employment used as independent variables in the left-most column. Estimated by the first-differenced GMM, using robust standard errors. Sales and value added are instrumented with lags at t − 3 and earlier. Employment is instrumented with lags at t − 4 and earlier. For year and industry dummies the joint F- statistic is reported. Standard errors in parentheses. P-values in brackets. The estimations on sales and value added correspond to equation (7). The estimation on employment corresponds to equation (9)

49 Page 51 of 53

ip t cr us

Table B22 Sensitivity of employment to temporary firm level shocks, 2004-2009.

Observations Hansen J-test

k=3

-0.0150* (0.00871) -0.0472*** (0.0177)

-0.0199** (0.00792) -0.0493*** (0.0159)

86,796 exactly identified

86,796 6.392 [0.0409]

k=1

k=2

k=3

-0.0193** (0.00791) -0.0562*** (0.0156)

-0.0178** (0.00747) -0.0811*** (0.0157)

-0.0171** (0.00739) -0.0804*** (0.0154)

-0.0186** (0.00732) -0.0819*** (0.0150)

86,796 11.066 [0.0258]

85,812 exactly identified

85,812 15.641 [0.0004]

85,812 17.856 [0.0013]

M

k=2

d

∆ǫit ×family firm

k=1

te

Temporary shocks ∆ǫit

an

∆ωit (dependent variable) Sales shocks Value added shocks (ǫ =sales residual) (ǫ =value added residual)

Ac ce p

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

Note: The dependent variable ∆ωit , is the first-differenced residual from the employment process in TableB21. Robust standard errors in parentheses. P-values in brackets. ǫ = are the residuals from the first stage estimations on sales and value added in Table B21. Instruments are defined as (∆ǫi,t+1 )k for k = 1, 2, 3. The feasible efficient GMM procedure is used for all estimations. *** p< 0.01, ** p< 0.05, * p< 0.1

50 Page 52 of 53

*Highlights (for review)

Highlights

ce pt

ed

M

an

us

cr

ip t

I examine the sensitivity of employment to shocks in family firms. I investigate fluctuations at the industry level and at the firm level. Family firms are less sensitive to performance and sales fluctuations Family firms might be more able to offer implicit employment protection.

Ac

   

Page 53 of 53