employment protection legislation, adjustment costs …dbyzalov/oecd.pdf · employment protection...
TRANSCRIPT
1
Employment Protection Legislation, Adjustment Costs
and Cross-country Differences in Cost Behavior
Rajiv D. Banker* Dmitri Byzalov† Lei (Tony) Chen‡
Abstract: Central to the economic theory behind sticky costs is the proposition that
managers take into account adjustment costs when changing resource levels. Few studies
have been able to test this theoretical prediction empirically due to the difficulty of
measuring adjustment costs directly. Adjustment costs associated with firing employees
that arise due to employment protection legislation (EPL) in different countries provides
an apt setting to test this prediction because prior research in economics provides strong
support for linking EPL with adjustment costs for labor. Using a sample of 15,833 firms
in 19 OECD countries during 1990-2008, we test the association between firm-level cost
stickiness and country-level EPL measures. We find that the degree of cost stickiness is
increasing in the strictness of EPL provisions, consistent with our hypotheses based on
adjustment costs, and supporting the theory that sticky cost behavior is driven by
managers’ deliberate resource commitment decisions in the presence of adjustment costs.
Keywords: Sticky costs; Employment protection legislation; Labor adjustment costs.
* Fox School of Business, Temple University, 461 Alter Hall, Philadelphia, PA 19122. E-Mail: [email protected]. Phone: (215) 204-2029. Fax: (215) 204-5587. † Fox School of Business, Temple University, 452 Alter Hall, Philadelphia, PA 19122. E-Mail: [email protected]. Phone: (215) 204-3927. Fax: (215) 204-5587. ‡ Guanghua School of Management, Peking University, Beijing, 100871. E-mail: [email protected]. Phone: (8610) 62753707. Fax: (8610)62756991.
2
1. Introduction
Recent studies have documented strong evidence of asymmetric cost behavior. Costs
are sticky when the cost response to a decrease in activity is smaller than the cost
response to an equivalent increase in activity (Anderson et al., 2003, Balakrishnan et al.,
2004, Banker et al., 2010b, Weiss 2010). The prevalence of cost stickiness calls into
question the validity of the traditional cost model which implies a mechanical symmetric
relationship between changes in activity and changes in costs. Recognizing this potential
source of asymmetry in cost behavior and hence in earnings changes has also been shown
to be informative in forecasting earnings and understanding earnings management in
financial accounting research (e.g., Banker and Chen, 2006, Weiss, 2010).
Prior research suggests that the key to understanding sticky cost behavior is to view
many costs as arising from deliberate resource commitment decisions made by managers,
and speculates that adjustment costs play a central role in these decisions (e.g., Anderson
et al., 2003). Prior research also informally describes the basic tradeoff that managers
would face in the presence of adjustment costs. In particular, in deciding how much to cut
committed resources when activity levels drop, managers would weigh the benefits of
more efficient operations against the adjustment costs to be incurred. This would lead
them to deliberately retain some of the underutilized resources in order to save on the
adjustment costs. At the same time, managers have much less discretion about acquiring
required resources when activity increases. Thus, to the extent that managers recognize
the role of adjustment costs and the associated tradeoffs, adjustment costs are expected to
moderate the extent of resource reductions for activity decreases without a commensurate
effect for activity increases, leading to cost stickiness (Anderson et al., 2003).
While the sticky costs literature alludes to the potential role of adjustment costs and
deliberate managerial decisions, the issue of optimal decisions in the presence of
adjustment costs has been explored in much greater depth in the literature on dynamic
factor demand in economics (e.g., Hamermesh, 1990, Bentolila and Bertola, 1990,
Caballero, 1991, Abel and Eberly, 1994, Hamermesh and Pfann, 1996, Dixit, 1997,
Eberly and Van Mieghem, 1997, Palm and Pfann, 1997, Goux et al., 2001). This
economics literature explicitly models the dynamic optimization problem faced by
3
forward-looking firms in the presence of adjustment costs. As we discuss in section 2, the
key tradeoff in this dynamic optimization is between the adjustment costs incurred for a
marginal unit of committed resources and the present value of net cash flows this
marginal unit is expected to generate over its useful life with the firm (e.g., Bentolila and
Bertola, 1990, Abel and Eberly, 1994). In other words, the firm will optimally increase
committed resources as long as the expected net present value for each added unit of
resources outweighs the upward adjustment costs. Conversely, it will optimally reduce
committed resources as long as the expected net present value for each eliminated unit of
resources is negative and large enough (in absolute value) to outweigh the downward
adjustment costs.
Besides providing a more nuanced view of the dynamic tradeoffs involved in
resource commitment decisions, the dynamic factor demand literature also shows that
asymmetry in the adjustment costs leads to asymmetry in the optimal decision rules. For
example, Caballero (1991) points out that “in general, for the asymmetric case, the stock
of capital responds more to ‘good’ than to ‘bad’ realizations” (page 284). Bentolila and
Bertola (1990) obtain similar analytical predictions for labor. Thus, the dynamic factor
demand literature in economics explicitly derives the cost accounting notion of cost
stickiness as a direct consequence of forward-looking optimal decisions in the presence
of asymmetric adjustment costs.1
The economic theory of optimal resource commitment decisions in the presence of
adjustment costs provides a plausible and theoretically sound potential explanation for
the widely-documented empirical patterns of cost stickiness. We will term this potential
explanation for cost stickiness the “economic theory of sticky costs.” However, there are
other plausible explanations for cost stickiness. For example, stickiness may arise for
purely mechanical reasons, since managers can reduce activity levels without cutting the
unused resources but they cannot similarly increase activity levels without acquiring the
required additional resources. Such physical asymmetry may lead to stickiness even in
the absence of any adjustment costs. Alternatively, stickiness may arise due to managers’
1 The dynamic factor demand literature focuses primarily on the implications of adjustment costs for macro outcomes like aggregate employment levels and aggregate business cycle dynamics. The implications for cost stickiness are mostly a useful by-product that has not been emphasized in this literature.
4
empire-building behavior (Chen et al., 2011). As we discuss in section 2, while any of
these mechanisms can generate cost stickiness in the data, they have different practical
implications. Therefore, it is important to directly examine the underlying fundamental
mechanism behind cost stickiness.2
In this paper, we test the central observable implication of the economic theory of
sticky costs. As we show in section 2, if cost stickiness reflects deliberate resource
commitment decisions by forward-looking managers who recognize the tradeoffs arising
due to adjustment costs, then the degree of cost stickiness should be increasing in the
magnitude of downward adjustment costs. On the other hand, alternative explanations for
cost stickiness do not imply any systematic relationship between adjustment costs and the
degree of cost stickiness. This allows us to test the economic theory of sticky costs
against alternative explanations.
Despite the importance of adjustment costs in the economic theory of sticky costs,
the relationship between adjustment costs and cost stickiness has yet to be tested
empirically. The lack of large-sample empirical evidence on this issue is primarily due to
the difficulty in measuring adjustment costs. Unlike costs actually incurred to provide
productive capacity, adjustment costs are typically opportunity costs not recorded in the
accounting system. Thus, direct measurement is hardly feasible, if not impossible, for
researchers. Moreover, it is not easy to find suitable research settings in which broad and
reliable proxies for the magnitude of adjustment costs are readily available.3
In this study, we exploit the provisions of employment protection legislation (EPL)
as a source of considerable adjustment costs for labor. As previous studies in labor
economics have demonstrated, EPL imposes substantial firing costs on firms (e.g., Long
and Siebert, 1983, Pissarides, 1999). This allows us to use indexes of EPL strictness,
2 There have been some recent claims that the findings of cost stickiness in the literature may be spurious findings that do not reflect actual asymmetries in cost behavior. Banker et al. (2010a) show that because of methodological errors in these studies their claims are unfounded. 3 Empirical studies of dynamic factor demand in economics (e.g., Bentolila and Bertola, 1991, Palm and Pfann, 1997, Goux et al., 2001) do not even attempt to find any observable proxies for adjustment costs. Instead, they assume that the observed resource adjustment decisions are optimal within a dynamic optimization model with adjustment costs, and back out the shape and magnitude of adjustment costs based on this assumption. This approach does not allow them to test the underlying theory of dynamic optimization with adjustment costs.
5
which are compiled and reported for most OECD countries, as reliable exogenous
empirical proxies for the adjustment costs associated with firing workers.
We develop hypotheses on the relationship between the degree of cost stickiness and
EPL strictness in different OECD countries. Specifically, we predict that, under the
economic theory of sticky costs, a firm operating in a country with more stringent EPL
provisions (i.e., greater downward adjustment costs for labor) would exhibit a greater
degree of cost stickiness, i.e., a greater degree of asymmetry in cost response to increases
and decreases in sales.
We conduct empirical tests of the relationship between EPL strictness (our empirical
proxy for labor adjustment costs) and cost stickiness using a large sample of publicly
listed companies from 19 OECD countries. The empirical results support our hypotheses
and are consistent with the economic theory of sticky costs. In other words, the
relationship between cost stickiness and EPL strictness that we observe in the data is
consistent with the theory that cost stickiness reflects the outcomes of deliberate resource
commitment decisions made by managers who recognize the dynamic tradeoffs that arise
because of adjustment costs.
We contribute to the growing literature on cost behavior by demonstrating that the
degree of cost stickiness varies across countries as a function of the strictness of EPL – a
widely-available empirical measure that has been shown in prior research in economics
(e.g., Long and Siebert, 1983, Lazear, 1990, Pissarides, 1999, Blanchard and Portugal,
2001) to be a reliable exogenous proxy for the adjustment costs which are central to the
economic theory of sticky costs. Our results show that a full understanding of cost
behavior in general and of cost stickiness in particular requires careful analysis not only
of the firm-specific factors analyzed in prior literature but also of the economy-wide
structural forces that shape managers’ resource adjustment decisions.
This paper is structured as follows. In section 2, we describe the economic theory of
sticky costs and employment protection legislation, and derive the empirical hypotheses.
In section 3, we describe the data and the empirical models. Section 4 presents the
empirical results, and section 5 concludes.
6
2. Sticky costs and employment protection legislation
While the traditional textbook view of cost behavior implies a symmetric mechanical
relationship between changes in activity and changes in costs, recent research on sticky
costs (e.g., Anderson et al., 2003, Balakrishnan et al., 2004, Banker et al., 2010a, Weiss,
2010) documents pervasive patterns of asymmetric cost behavior (stickiness and anti-
stickiness4) which are inconsistent with the traditional view.
In an effort to explain the source of such asymmetries in cost behavior, the sticky
costs literature suggests that many costs arise as a result of deliberate resource
commitment decisions made by managers. When managers adjust committed resources,
they have to incur adjustment costs such as hiring and firing costs for labor, or
installation and disposal costs for equipment. The sticky costs literature speculates that
adjustment costs play a central role in cost behavior, and alludes to the dynamic tradeoffs
that managers would face in the presence of adjustment costs. In particular, in deciding
how much to adjust committed resources in response to a given contemporaneous change
in activity, managers would weigh the profit consequences of carrying too much or too
little capacity against the adjustment costs that would have to be incurred when changing
the committed resources in the current period and in the future (e.g., Anderson et al.,
2003). To the extent that managers recognize this tradeoff in their decisions, it would
introduce more complex dynamics in cost behavior,5 giving rise to patterns of stickiness
and anti-stickiness.
While the sticky costs literature informally describes some of the dynamic
considerations that arise because of adjustment costs, this dynamic tradeoff has been
modeled formally in the literature on dynamic factor demand in economics (e.g.,
Hamermesh, 1990, Bentolila and Bertola, 1990, Caballero, 1991, Abel and Eberly, 1994,
4 Formally, we say that costs are “sticky” if they increase more for a one-percent increase in activity than they decrease for an equivalent decrease in activity. Conversely, costs are “anti-sticky” (Weiss, 2010) if they increase less for a one-percent increase in activity than they decrease for an equivalent decrease in activity. 5 Specifically, managers’ resource adjustment decisions would be driven not only by the contemporaneous changes in activity, but also by the level of capacity carried over from the prior period (since it affects the amount of adjustment costs that would have to be incurred to achieve the desired capacity level in the current period) and by managers’ expectations for future activity levels (since they affect the amount of adjustment costs that would have to be incurred to achieve the desired levels of capacity in the future).
7
Hamermesh and Pfann, 1996, Dixit, 1997, Eberly and Van Mieghem, 1997, Palm and
Pfann, 1997, Goux et al., 2001). This economics literature explicitly models the dynamic
optimization problem faced by forward-looking firms in the presence of adjustment costs.
At the optimal level of committed resources in this dynamic optimization, the marginal
adjustment costs incurred per unit of committed resources in the current period should be
equal to the present value of expected net cash flows generated by the marginal unit of
committed resources over its useful life with the firm (Bentolila and Bertola, 1990, Abel
and Eberly, 1994). For example, in the context of optimal hiring and firing decisions, the
present value of net cash flows generated by a marginal worker consists of the marginal
revenue product she is expected to generate over her tenure with the firm net of her
wages over the same period, and net of the expected future firing costs that will be
incurred at the (random) end of her tenure, all discounted to the present period (Bentolila
and Bertola, 1990). The firm will optimally hire additional workers as long as the present
value of expected net cash flows of the marginal worker exceeds the hiring costs (upward
adjustment costs per worker). Conversely, the firm will optimally fire workers if the
present value of net cash flows of the marginal worker is negative and large enough (in
absolute value) to exceed the firing costs (downward adjustment costs). In other words, it
will fire workers only if it is costlier for the firm to keep them than to fire them.
Furthermore, after the firm has fired the optimal amount of workers, the net present value
of the marginal surviving worker will still be negative (equal to the firing cost with a
minus sign), i.e., the firm will optimally retain some of its underutilized workers in order
to save on the firing costs.
One key insight from the dynamic factor demand literature is that asymmetry in the
adjustment costs translates into asymmetry in the optimal resource commitment decisions
(e.g., Bentolila and Bertola, 1990, Caballero, 1991). For example, in the context of labor
resource adjustment decisions, if the firing costs per worker are larger than the hiring
costs, then the firm will be more reluctant to fire workers when demand drops than to hire
workers when demand increases by an equivalent amount. In particular, the firm will hire
additional workers only if the favorable demand shock is strong enough to increase the
net present value of its marginal worker to above the hiring threshold, equal to the hiring
cost per worker. Likewise, the firm will fire workers only if the adverse demand shock is
8
strong enough to reduce the net present value of its marginal worker to below the firing
threshold – a negative number equal to minus the firing cost per worker.6 If the firing
costs are larger than the hiring costs, the adverse demand shocks that trigger the firing
decision would have to be relatively more severe than the favorable demand shocks that
trigger the hiring decision. In other words, the firm will be less likely to fire workers for
activity decreases than to hire workers for equivalent activity increases. Thus, the
dynamic factor demand literature in economics explicitly derives the cost accounting
notion of cost stickiness as a direct consequence of forward-looking optimizing behavior
in the presence of asymmetric adjustment costs.
The economic theory of optimal resource commitment decisions in the presence of
adjustment costs provides a plausible and theoretically sound potential explanation for
the widely-documented empirical patterns of cost stickiness. We will term this potential
explanation for cost stickiness the “economic theory of cost stickiness.” However, there
are other plausible explanations for cost stickiness. For example, stickiness may arise for
purely mechanical reasons, since it is physically impossible to increase activity levels
without acquiring the required additional resources, but at the same time it is physically
possible to reduce the activity levels without cutting the unused resources. Furthermore,
cutting the unused resources requires substantial managerial discretion and effort (e.g.,
Anderson et al., 2003). This may lead to stickiness, even in the absence of any adjustment
costs, if managers do not exercise sufficient discretion and effort in identifying and
eliminating the unused resources. Alternatively, stickiness may arise due to managers’
empire-building behavior (Chen et al., 2011), since empire-building managers will be
eager to expand the resources under their control when activity increases but reluctant to
cut back unused resources when activity declines. This may also lead to cost stickiness
even in the absence of adjustment costs.
While any of these explanations may lead to cost stickiness in the data, they have
very different implications. In particular, if stickiness arises due to mechanical reasons or
due to empire-building behavior, it would reflect the consequences of wasteful
6 With linear hiring and firing costs, there is also a region of inaction. Specifically, if the net present value of the marginal worker is above the firing threshold and below the hiring threshold, then the firm will not adjust its work force.
9
managerial behavior that reduces firm value. On the other hand, under the economic
theory of sticky costs (i.e., dynamic optimization by managers in the presence of
adjustment costs), the same observed pattern of stickiness would represent the
consequences of desirable managerial behavior that increases firm value. Therefore, it is
important to examine the underlying fundamental mechanism behind cost stickiness.7
In this paper, we exploit the central testable implication of the economic theory of
sticky costs in order to test it against alternative explanations. As we discuss next, the
economic theory of sticky costs implies a systematic positive relationship between the
magnitude of downward adjustment costs and the degree of cost stickiness, which can be
tested using appropriate empirical proxies for the magnitude of adjustment costs. At the
same time, alternative explanations mentioned above do not imply any clear relationship
between the magnitude of adjustment costs and the degree of cost stickiness.
In deriving our predictions, we build on prior literature on dynamic factor demand in
economics (e.g., Bentolila and Bertola, 1990, Caballero, 1991, Abel and Eberly, 1994).
While this prior literature has focused primarily on the macroeconomic implications of
adjustment costs for aggregate employment levels and business cycle dynamics, we
leverage their insights to generate our predictions for firm-level resource commitment
decisions made by managers and their implications for firm-level cost behavior. For
clarity of exposition, we describe the relationship between downward adjustment costs
and stickiness in the context of labor resources, where the adjustment costs are the firing
and hiring costs per employee (the intuition for other capacity resources is similar). When
the firing costs per employee are higher, it affects the hiring and firing decisions through
two channels. First, conditional on the original level of the work force carried over from
the prior period, higher firing costs moderate the magnitudes of both the hiring decisions
for activity increases and the firing decisions for activity decreases. As we explain below,
this moderating effect of higher firing costs is disproportionately larger for activity
decreases than for activity increases. As a result, conditional on the original level of the
7 The underlying mechanism behind cost stickiness also has important implications for the design of managers’ incentives and performance evaluation. For example, if stickiness arises due to mechanical reasons or empire-building behavior, then the optimal incentive contract would involve penalizing managers for retaining slack resources in a downturn. On the other hand, under the economic theory of sticky costs, the same incentive contract would reduce firm value by forcing managers to reduce slack resources below the optimal level from the dynamic optimization perspective.
10
work force, higher firing costs increase the degree of cost stickiness. Second, higher
firing costs affect the typical level of the work force (and hence the typical level of slack
labor) carried over from the prior period. As we explain below, higher firing costs reduce
the average amount of slack labor carried over from the prior period, which further
amplifies cost stickiness.
The intuition for the first channel is as follows. In its hiring decisions, the firm will
keep adding workers as long as the net present value of the marginal worker is positive
and above the hiring cost per worker (e.g., Bentolila and Bertola, 1990, Goux et al., 2001).
Similarly, in its firing decisions, the firm will keep laying off workers as long as the net
present value of the marginal worker is negative and below the firing cost per worker
(with a minus sign8). In other words, the firm will fire workers if the cost of keeping them
(i.e., negative present value) exceeds the cost of firing them. When the firing costs per
worker are higher, it reduces the net present value of the marginal worker by making it
costlier for the firm to lay off this worker in the future if necessary.9 In other words, by
increasing the anticipated cost of future layoffs, higher firing costs make the marginal
worker less valuable for the firm, both in the context of hiring decisions and in the
context of firing decisions (e.g., Bentolila and Bertola, 1990). As a result, for the same
increase in activity (and for the same original level of the work force), the firm will
optimally hire fewer additional workers, since the (unchanged) hiring costs per worker
are now traded off against lower net present value generated by this worker. In the
context of firing decisions, higher firings costs have a similar negative effect on the net
present value of the marginal worker, making it less desirable for the firm to keep the
marginal worker. However, unlike in the hiring context, higher firing costs have an
additional direct effect of making it even less desirable for the firm to lay off the marginal
worker. The latter effect will dominate since the firing costs have to be incurred
immediately and with certainty when laying off workers, while the anticipated future
firing costs embedded in the net present value of the marginal worker will be incurred
only in the future (and therefore are discounted), and they will be incurred only if future
8 The sign is reversed for firing decisions since the firm gives up the net present value of the fired workers. 9 Recall that the net present value of the marginal worker consists of all expected net cash flows during her tenure with the firm, including the firing costs to be incurred at the (random) future end of her tenure, all discounted to the present period (e.g., Bentolila and Bertola, 1990).
11
shocks turn out to be sufficiently adverse to necessitate layoffs (and not with certainty).
As a result, for the same decrease in activity (and for the same original level of the work
force), the firm will optimally lay off fewer workers when the firing costs are higher,
since the costs associated with firing workers increase to a greater extent than the costs
(negative present value) associated with retaining underutilized workers. Thus, for a
given original level of the work force carried over from the prior period, higher firing
costs will moderate both the hiring decisions for activity increases and the firing
decisions for activity decreases, and this moderating effect will be disproportionately
stronger for the firing decisions.10 This will increase the degree of cost stickiness (or
reduce the degree of anti-stickiness in situations in which labor is anti-sticky11).
Second, higher firing costs will affect the typical levels of slack labor carried over
from the prior period, which will further affect the degree of cost stickiness (e.g.,
Balakrishnan et al., 2004). Firing costs have two offsetting effects on the amount of slack.
On the one hand, by moderating the firing decisions for activity decreases, higher firing
costs increase the amount of slack labor (e.g., Bentolila and Bertola, 1990). On the other
hand, by moderating the hiring decisions for activity increases, higher firing costs
decrease the amount of slack labor. On average, we expect the latter effect to dominate
since activity increases are much more common in the data than activity decreases (64.1
percent increases vs. 35.9 percent decreases in our sample). This will result in less slack
carried over from the prior period on average. In turn, this will further moderate the firing
decisions for activity decreases, since there are fewer underutilized workers to begin with.
At the same time, less slack will amplify the hiring decisions for activity increases, since
managers will have to hire more additional workers to accommodate a given activity
increase. Consequently, by reducing the average amount of slack, higher firing costs will
further increase the degree of cost stickiness (or reduce the degree of anti-stickiness in
situations in which labor is anti-sticky).
Summing up, under the economic theory of sticky costs, higher firing costs will
increase the degree of cost stickiness (or reduce the degree of anti-stickiness) for labor,
10 Bentolila and Bertola (1990) derive this prediction formally in an analytical dynamic model of optimal labor demand with firing and hiring costs. 11 Banker et al. (2010b) show that costs are likely to be anti-sticky in periods when managers’ expectations for future activity levels are sufficiently pessimistic.
12
via both channels discussed above. The same intuition applies more generally for any
type of capacity resources that are costly to adjust. Higher downward adjustment costs
will increase the degree of cost stickiness (or reduce the degree of cost anti-stickiness),
while higher upward adjustment costs will reduce the degree of stickiness (or increase the
degree of anti-stickiness). This implication is central to the economic theory of sticky
costs, and adjustment costs are one of the key determinants of cost behavior if this theory
is valid.
At the same time, alternative explanations for cost stickiness do not imply any
systematic relationship between the magnitude of adjustment costs and the degree of cost
stickiness. For example, if cost stickiness arises for mechanical reasons due to managers’
failure to identify and eliminate unused resources, we would not expect any systematic
relationship between the magnitude of adjustment costs and the degree of managerial
inattention that is driving stickiness under this explanation. Similarly, if cost stickiness
arises due to managers’ empire-building behavior, we would not expect any systematic
relationship between the magnitude of adjustment costs and the intensity of empire-
building incentives. This difference allows us to test the observable implications of the
economic theory of sticky costs against alternative explanations.
Despite the importance of adjustment costs in the economic theory of sticky costs,
few studies have been able to test the relationship between adjustment costs and cost
stickiness empirically. The main complication in testing this relationship is that
adjustment costs are hard to measure directly. As Hamermesh and Pfann (1996) point out,
many of the adjustment costs are implicit costs of lost output,12 not explicit monetary
costs captured in the accounting system. Despite this complication, several studies have
been able to test this relationship by relying on observable firm-level proxies for the
magnitude of adjustment costs such as asset and employee intensity (Anderson et al.,
2003) or sales volatility (Kama and Weiss, 2010), and by linking these proxies to firm-
level variation in cost stickiness. In contrast, in this paper we exploit country-level
proxies for adjustment costs for labor, based on the strictness of employment protection
12 For example, after firing workers, the firm may have to reassign some of the surviving workers to new tasks, temporarily reducing their productivity. Likewise, after hiring new workers, more experienced (and more productive) workers may have to spend time training them, resulting in lower productivity. Similarly, productivity may decline temporarily if the firm added new equipment or got rid of its existing equipment.
13
legislation in each country, and we link them to cross-country variation in cost stickiness
(in addition, we also control for firm-level drivers of cost stickiness following prior
literature). Compared to proxies for adjustment costs used in prior literature such as asset
intensity or sales volatility, one key advantage of employment protection measures is that
they are exogenous with respect to managers’ resource commitment decisions.13
Employment protection legislation (EPL), our empirical proxy for labor adjustment
costs, is a central part of firms’ institutional environment. EPL describes rules about
dismissal of employees, such as restrictions on firing after some period of service, the
compulsory payment of redundancy payments and the length of notice required before
dismissal. Acting as a tax on dismissals, employment protection imposes substantial
firing costs on employers (Long and Siebert, 1983, Pissarides, 1999, OECD, 2004).14 As
we document in section 3, there are large differences in EPL strictness across the
countries in our data, providing a rich source of exogenous cross-country variation in the
firing costs.
Prior research in labor economics has documented extensively that EPL plays a
substantial role in macroeconomic outcomes such as labor force participation, job
creation and destruction, unemployment and productivity growth. Lazear (1990),
Bentolila and Bertola (1990), Hopenhayn and Rogerson (1993), Mortensen and Pissarides
(1999), Heckman et al. (2000), Blanchard and Portugal (2001), Besley and Burgess
(2004), Botero et al. (2004), Boeri and Jimeno (2005), Haltiwanger et al. (2006), Micco
and Pages (2006), Autor et al. (2007), Kugler and Pica (2008), Schivardi and Torrini
(2008) document large significant effects of EPL strictness on the long-run aggregate
employment levels and other labor market outcomes, for a wide variety of developed and
developing countries. DeFreitas and Marshall (1998), Besley and Burgess (2004), Micco
and Pages (2006), Pierre and Scarpetta (2006), Autor et al. (2007), Cingano et al. (2008)
and Bassanini et al. (2009) document significant effects of EPL strictness on long-term
productivity growth. Mortensen and Pissarides (1999), Burgess et al. (2000), Barone
13 In contrast, measures like asset intensity or sales volatility are in part outcomes of prior managerial decisions in response to shocks that may still have a persistent direct effect on cost behavior in the current period. 14 For example, in Portugal, which has the strictest EPL in our data, mandated severance pay is equal to 3 months of regular pay for a worker with 9 months of tenure, and it increases to 20 months of regular pay for a worker with 20 years of tenure (OECD, 2004).
14
(2001) and Caballero (2004) find that stricter EPL reduces firms' ability to adjust to
shocks.
Thus, prior research in labor economics documents that (1) EPL is a major
determinant of firing costs, and (2) it has important effects for a wide range of
macroeconomic outcomes. We exploit the same cross-country institutional variation in
EPL, however our main focus is different: we are interested in the role of EPL in firm-
level cost behavior (which is of interest to firm managers in planning resource levels and
costs) rather than its role in macroeconomic outcomes (which is of interest for policy-
makers in formulating broad macroeconomic policy). Despite the different focus of our
study, we are able to leverage the key insights from these previous studies regarding the
impact of EPL on adjustment costs to develop our empirical hypotheses.
As we discuss above, the economic theory of sticky costs implies that higher
downward adjustment costs should lead to more stickiness in resource adjustment. Since
stricter EPL increases the magnitude of firing costs (downward adjustment costs for
labor), we expect it to increase the degree of stickiness for labor costs, and, since direct
and indirect labor costs account for a large fraction of operating costs, we expect it to
increase the degree of stickiness for operating costs.15 This leads us to our first empirical
hypothesis:
Hypothesis 1: Stricter employment protection legislation (EPL) is associated with a
higher degree of cost stickiness for operating costs.
2.1. Differential effects of EPL for regular and temporary employees
Overall employment protection legislation provisions consist of two distinct
components: employment protection for regular employees and regulation of temporary
forms of employment (see Table 1 and section 3 for details). As we show in section 3,
there are significant cross-country differences in relative employment protection for
15 We focus on operating costs, as opposed to more detailed measures of labor costs, due to data limitations for labor costs in Compustat. Specifically, labor costs have a high proportion of missing data (69 percent missing in our full sample), and even when data is available, it is less reliable than data on operating costs. Among firms that report labor costs, they account for 28 percent of total operating costs on average.
15
temporary vs. regular employees, i.e., some countries have strict EPL for regular but not
for temporary employees and vice versa.
Both forms of EPL are likely to affect the magnitude of downward adjustment costs
for labor. In particular, EPL for regular workers specifies the severance pay levels, which
gives rise to explicit monetary costs associated with firing regular workers (e.g., OECD,
2004). It also specifies the regulations and procedures that constrain employer’s ability to
fire workers, such as notice length and notification procedures before dismissal or the
rules related to unfair dismissal, which give rise to additional, non-monetary, firing costs.
EPL for temporary employees also affects labor adjustment costs, but through a different
channel. In particular, it constrains employers’ ability to use temporary workers (who
have low firing costs) instead of regular workers (who have much higher firing costs).
Thus, stricter EPL for temporary employees increases the effective monetary and non-
monetary firing costs that employers face, by shifting the composition of their work force
from temporary to regular employees.
Stricter EPL for regular workers and stricter EPL for temporary workers both
increase the downward adjustment costs for labor. Therefore, based on the economic
theory of sticky costs, we expect both forms of EPL to increase the degree of cost
stickiness for labor costs, and, since labor costs represent a large fraction of total
operating costs, we expect them to increase the degree of stickiness for operating costs.
This leads us to the following hypotheses:
Hypothesis 2a: Stricter employment protection for regular employees is associated
with a higher degree of cost stickiness for operating costs.
Hypothesis 2b: Stricter employment protection for temporary employees is
associated with a higher degree of cost stickiness for operating costs.
The relative magnitudes of the effects of regular and temporary EPL on cost
stickiness are a priori ambiguous. For example, companies may use temporary employees
in all areas of operations, in which case they will be able to accommodate most of the
fluctuations in their labor requirements by adjusting the number of temporary workers,
16
taking advantage of the much lower firing costs for them relative to regular workers. In
this case, even though temporary employees account for a relatively small fraction of the
total work force, they would account for most of the changes in individual companies’
work force. If so, EPL for temporary workers would have a much greater effect on the
degree of cost stickiness than EPL for regular workers. Conversely, the use of temporary
workers may be confined to a relatively narrow subset of operations (for example,
generic tasks that can be accomplished without much firm-specific human capital and
experience). If so, labor force adjustments would involve hiring and firing regular
workers alongside temporary workers, in which case EPL for regular workers might have
a much stronger effect on cost stickiness than EPL for temporary workers. Thus, a priori
either form of EPL (regular or temporary) may have a dominant effect on cost stickiness.
3. Research methodology
We empirically examine the relationship between country-specific employment
protection legislation and cost stickiness for firms in OECD member countries.16 We
choose this research setting primarily for two reasons: (1) OECD includes all the major
industrialized nations with a free market economy, and (2) measures of country-specific
EPL and other labor market characteristics are reliably and systematically reported. In
addition, as discussed in the previous section, there is a rich literature in labor economics
that has examined various aspects of EPL and other labor market characteristics for
OECD countries (e.g., Lazear, 1990, Bentolila and Bertola, 1990, Hopenhayn and
Rogerson, 1993, Mortensen and Pissarides, 1999, Heckman et al., 2000, Botero et al.,
2004, and many others). Since our main goal in this study is to explore the relationship
between EPL and cost behavior, we are able to leverage this literature both in formulating
our empirical hypotheses and in identifying the appropriate empirical measures of EPL
and additional labor market control variables.
16 At the beginning of our sample period (1990), OECD consisted of 24 member countries. Six more countries joined OECD during the sample period 1990-2008 (Czech Republic, Hungary, Mexico, Poland, South Korea and Slovakia), and four more countries joined after 2008 (Chile, Estonia, Israel and Slovenia).
17
3.1. Measures of employment protection legislation
To measure the strictness of employment protection legislation for regular and
temporary employees, we use indexes defined and reported in OECD (2004). These
indexes are based on the legislative provisions governing the firing of workers (such as
the length of the notice period before dismissal, or the level of severance pay), as well as
regulations related to temporary forms of employment (such as the maximum duration of
successive fixed-term contracts, or restrictions on employment through temp agencies).
For each country, OECD (2004) characterizes employment protection legislation for
regular and temporary employees along 14 basic items, and then combines them into
summary indexes of EPL strictness for regular employees REGEPLn and temporary
employees TEMPEPLn for country n (see Table 1 for details). These indexes are
normalized to range from 0 to 6, with higher scores representing stricter regulation.
Table 3 presents the EPL indexes for the 19 OECD countries in our main sample
(Australia, Austria, Belgium, Canada, Denmark, Finland, Germany, Ireland, Italy, Japan,
Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, UK and US).17 There is
substantial cross-country variation in EPL strictness for regular employees, with lowest
scores for the US, UK and Switzerland (0.2, 0.9 and 1.2 respectively on a scale from 0 to
6), and the highest scores for Portugal, Netherlands, Austria and Sweden (4.3, 3.1, 2.9
and 2.9 respectively).18 Countries with stricter EPL for regular workers typically have
stricter EPL for temporary workers as well (the correlation between the two EPL indexes
is 0.789), but there is also some meaningful independent variation in EPL for temporary
workers relative to regular workers. For example, France and Italy have moderate levels
of EPL for regular employees (1.8 and 2.3 respectively, vs. the median level of 2.3), but
they have stricter EPL for temporary employees than any other country in the sample (3.3,
vs. the median of 1.6). At the opposite extreme, the Netherlands has one of the strictest 17 We do not include countries that joined the OECD after the beginning of our sample period, since most of them are the transition economies in Eastern Europe. We also discard OECD countries with missing data on labor market characteristics (Iceland, Luxembourg, Mexico, New Zealand and Turkey). 18 To illustrate how substantial this variation is, we also examine the mandated severance pay regulations since they give rise to well-interpretable monetary firing costs. Mandated severance pay for a worker with 4 years of tenure ranges from zero in the US to 2.6 months of regular pay in Spain and 4 months of regular pay in Portugal. For a worker with 20 years of tenure, mandated severance pay ranges from zero in the US to 12 months of regular pay in Spain and 20 months of regular pay in Portugal (OECD, 2004).
18
levels of EPL for regular employees (3.1), but below-median levels of EPL for temporary
employees (1.2, vs. the median of 1.6).
3.2. Sample selection and descriptive statistics
We start by drawing the full sample of publicly-listed non-financial firms from
Compustat (Global and North America) between 1988-2008 for the 19 OECD countries
we focus on.19 We deflate sales and operating costs to control for inflation, using country-
specific GDP deflators. We discard firm-years if: (1) sales or operating costs are missing
or negative in current or two prior years, (2) operating costs are less than 50% or more
than 200% of sales in current or two prior years, or (3) assets are missing or negative in
current year. We also discard firms reporting in a non-native currency (for example,
European firms reporting in US dollars).20 After that, we discard 1% outliers on each tail
for the dependent variable (log-change in deflated operating costs) and for the continuous
firm-level explanatory variables (log-change in deflated sales and asset intensity). We
also discard firm-years if deflated sales increased by over 50% or dropped by over 33%
in current or prior year,21 since such extreme year-on-year changes in sales mostly reflect
mergers or divestitures. The final sample in estimation consists of 128,333 observations
for 15,833 firms in 19 OECD countries between 1990-2008.22
We merge this sample with the EPL data described earlier and additional country-
level control variables from several sources. The data on annual GDP growth rates and
inflation rates (GDP deflators) for each country is from the World Bank Databank.23 The
data on labor market control variables in robustness checks (union density, bargaining
coordination and centralization index and unemployment benefits index) is from OECD
(2004) and Nickell et al. (2005). The variable definitions in estimation are summarized in
Table 2. 19 For most OECD countries in Global Compustat, data is available only starting from 1988. For consistency, we start the sample in 1988 for all countries in our data. 20 The results are similar when we do not discard such firms. 21 These percentage cutoffs (-33% and +50%) are symmetric when transformed into the log-change form (ln(2/3) and ln(3/2) respectively) that we use in estimation. 22 The first two lags in the raw data are used up in computing the first differences and preparing the control variables, so the final sample in estimation starts in 1990 rather than in 1988. 23 http://databank.worldbank.org/ddp/home.do
19
The descriptive statistics are presented in Table 3. Besides large differences in EPL
discussed in the previous subsection, there are important cross-country differences in
other variables. For example, average annual GDP growth ranges from 1.1 percent in
Italy and 1.3 percent in Japan to 6.2 percent in Ireland. The average annual log-change in
deflated sales ranges from 0.022 (2.2 percent a year) for Japanese firms to 0.051 (5.1
percent) for Swedish firms and 0.053 (5.3 percent) for Irish firms. As Anderson et al.
(2003) and Banker et al. (2010b) show, managers’ optimism and pessimism regarding
future sales play an important role in sticky cost behavior. Thus, the cross-country
differences in growth rates are likely to lead to large cross-country differences in cost
stickiness due to differences in managers’ expectations. Therefore, in the empirical
analysis, we include empirical proxies for managers’ optimism and pessimism following
Anderson et al. (2003) to control for cross-country and within-country differences in
managers’ expectations regarding future sales.
3.3. Empirical model
Our empirical estimation is based on a hierarchical linear model on which the
behavior of a level-1 outcome (i.e., firm-level cost behavior) is postulated to be a
function of level-2 explanatory variables (i.e., country-level characteristics) and firm-
level control variables. We next specify the level-1 and level-2 models and discuss the
model structure in detail.
We begin with the following firm-level model of cost behavior linking annual
changes in deflated operating costs (XOPR) to contemporaneous changes in deflated
sales revenue (SALE), which follows the sticky costs model of Noreen and Soderstrom
(1997) and Anderson et al. (2003)
tintintintintintintin uSALEDECSALEXOPR ,,,,,,,,,2,,,,,10,, lnlnln (1)
where ∆lnXOPRn,i,t represents the log-change in deflated operating costs for firm i in
country n in year t, ∆lnSALEn,i,t is the log-change in deflated sales, DECn,i,t is a binary
variable equal to one if deflated sales decreased in year t and zero otherwise, un,i,t is an
error term that has mean zero and is independent of the explanatory variables, and the
slopes α1,n,i,t and α2,n,i,t are specified in detail below. In this specification, the slope α1,n,i,t
20
approximates the percentage change in costs for a one percent increase in sales, and
α1,n,i,t+α2,n,i,t approximates the percentage change in costs for a one percent decrease in
sales. The cost stickiness coefficient α2,n,i,t captures the degree of asymmetry in cost
behavior (stickiness if α2,n,i,t is negative, and anti-stickiness if α2,n,i,t is positive).
We introduce the level-2 model by specifying the firm-level slopes α1,n,i,t and α2,n,i,t in
model (1) as a function of country-level explanatory variables, firm-level control
variables following prior studies (e.g., Anderson et al., 2003), and additional country-
level random effects. In the basic model (Model A), we specify the slope coefficients as
nnntintntin vEPLLAWAINTGDPGROWTH ,154,,3,21,,,1 (2a)
nnntintntintin vEPLLAWAINTGDPGROWTHDEC ,21110,,9,81,,76,,,2 (2b)
where GDPGROWTHn,t represents real GDP growth rate in country n in year t, AINTn,i,t
represents asset intensity (log ratio of total assets to sales) for firm i in year t, LAWn is a
binary variable equal to one for common-law countries, DECn,i,t-1 is a binary variable
equal to one if sales decreased in prior period, EPLn is the employment protection
legislation index for country n, and v1,n, v2,n are country-level random effects. Following
prior literature (e.g., Anderson et al., 2003), we use GDP growth (GDPGROWTHn,t) and
a dummy for successive decreases in sales (DECn,i,t-124 ) as empirical proxies for
managers’ optimism or pessimism regarding future sales, and we use asset intensity
(AINTn,i,t) as an empirical proxy for the magnitude of adjustment costs facing the firm.25
We include the common-law dummy LAWn (equal to one for Australia, Canada, Ireland,
UK and US), since prior research in economics (e.g., La Porta et al., 1997, 1998, 2000;
Djankov et al., 2007) has found that the legal origin of a country (common law vs. code
law) is one of the primary drivers of cross-country differences in corporate governance,
access to external finance, business regulation and other outcomes that likely play an
24 Since DECn,i,t-1 enters the slope only for current decreases in sales (DECn,i,t=1), it is relevant only for successive decreases taking place both in year t and in year t-1. 25 We do not include employee intensity as a control variable in our main specification, since the number of employees variable has a high proportion of missing observations, and since cross-country comparisons of employee intensity are contaminated by cross-country differences in labor productivity and fluctuations in market exchange rates. In untabled robustness checks, we get similar results after controlling for employee intensity.
21
important role in firm-level cost behavior.26 The country-level random effects v1,n, v2,n
capture the cross-country differences in cost behavior that are not accounted for by the
observable explanatory variables. By construction, v1,n, v2,n have mean zero and they are
independent of the explanatory variables in the regression. The main parameter of interest
in estimation is β11, which captures the relationship between the strictness of EPL and the
degree of cost stickiness. Hypothesis 1 implies that β11 should be negative, i.e., stricter
EPL should be associated with a higher degree of cost stickiness (a more negative
α2,n,i,t).27
By combining equation (1) with (2a) and (2b), we obtain our basic estimation model:
Model A
tintintinn
ntintntin
tinnntintntin
SALEDECEPLLAWAINTGDPGROWTHDEC
SALEEPLLAWAINTGDPGROWTHXOPR
,,,,,,11
10,,9,81,,76
,,54,,3,210,,
ln)(
ln)(ln
(3)
where εn,i,t is the error term, which combines the residuals from equations (1), (2a) and
(2b), and the rest of the terms have been described above. The error term εn,i,t can be
rewritten as
tintinntinntintin SALEDECvSALEvu ,,,,,2,,,1,,,, lnln (4)
where the original random shocks un,i,t, v1,n, v2,n from equations (1), (2a), (2b) have mean
zero and are independent of the explanatory variables in the regression. Consequently, the
combined error term εn,i,t in equation (3) has zero mean for any value of the explanatory
variables, and therefore ordinary least squares (OLS) yields unbiased and consistent
estimates of the parameters in equation (3). The inclusion of country-level random effects
v1,n, v2,n introduces cross-sectional correlation in εn,i,t across firms within each country. It
also introduces heteroskedasticity, since the random shocks v1,n, v2,n are multiplied by
∆lnSALE and DEC×∆lnSALE respectively. To deal with these issues, we use clustered
standard errors (Rogers, 1993) with clustering by country.28
26 Calleja et al. (2006) also find (for a limited sample of four developed countries) that legal origin plays a role in cost behavior. 27 We also control for EPL strictness in the slope for increases α1,n,i,t since EPL strictness is likely to affect firms’ hiring decisions for activity increases. However, as we discuss in section 2, the expected sign of this effect is a priori ambiguous. 28 Alternatively, instead of using OLS with appropriately clustered standard errors, we could estimate Model A as a random-coefficients model using maximum likelihood. However, this approach would be less
22
We also estimate an extended model (Model B), in which we replace the aggregate
EPL index (EPLn) with two more detailed indexes, which measure the strictness of EPL
separately for regular and temporary employees. The slope coefficients in Model (B) are
specified as:
nnnntintntin vTEMPEPLREGEPLLAWAINTGDPGROWTH ,1654,,3,21,,,1 (5a)
nnn
ntintntintin
vTEMPEPLREGEPLLAWAINTGDPGROWTHDEC
,21312
11,,10,91,,87,,,2
(5b)
where REGEPLn represents the index of regular employment protection in country n,
TEMPEPLn is the index of temporary employment regulation in country n, and the rest of
the terms have been described above. The main parameters of interest in estimation are
β12 and β13. Hypotheses 2a, 2b imply that β12 and β13 should both be negative, i.e., an
increase in strictness of EPL for regular or temporary employees should be associated
with a higher degree of cost stickiness (a more negative stickiness coefficient α2,n,i,t).29
Proceeding similarly to Model A above, we obtain the estimation equation for Model
B by combining equation (1) with equations (5a), (5b):
Model B
tintintinnn
ntintntin
tinnn
ntintntin
SALEDECTEMPEPLREGEPLLAWAINTGDPGROWTHDEC
SALETEMPEPLREGEPLLAWAINTGDPGROWTHXOPR
,,,,,,1312
11,,10,91,,87
,,65
4,,3,210,,
ln)(
ln)(ln
(6)
We estimate Model B using OLS with standard errors clustered by country.
In robustness checks, we introduce additional labor market control variables (union
density, collective bargaining coordination and centralization index, and unemployment
benefits index) in the slope coefficients α1,n,i,t, α2,n,i,t. The final estimation models in these
robust, since it would require us to impose additional distributional assumptions on the error terms in equations (1) and (2a), (2b), including their time-series structure, and it would yield consistent estimates only if these additional distributional assumptions hold in the data. When we impose these additional assumptions and estimate Model A as a random-coefficients model using maximum likelihood, the main results are similar. 29 We also control for REGEPLn and TEMPEPLn in the slope for increases α1,n,i,t since these variables are likely to affect firms’ hiring decisions for increases in sales. As we discuss in section 2, the expected sign of this effect is a priori ambiguous.
23
robustness checks are analogous to Models A and B above, with additional interaction
terms for the labor market controls.
4. Empirical results
We present the main estimation results for Model A in column (a) of Table 4. The
coefficients on the control variables (asset intensity, GDP growth and a dummy for
successive decreases in sales) have the expected signs and are consistent with the
findings in prior literature for the US data (e.g., Anderson et al., 2003). The main
parameter of interest is β11, which captures the association between the strictness of
employment protection legislation EPLn and the degree of cost stickiness. The estimate is
β11=-0.044 (t=-4.03), negative and significant at the 1 percent level. Thus, stricter EPL in
our data is associated with a higher degree of cost stickiness (i.e., a more negative
stickiness coefficient α2,n,i,t), lending strong support to Hypothesis 1.
In addition to being statistically significant, the magnitude of β11 is also
economically significant. For example, if we compare Switzerland (the country with the
least strict EPL in continental Europe) and Portugal (the country with the strictest EPL in
continental Europe), the predicted cost stickiness coefficient for a firm in Portugal is 0.11
higher (in absolute value) than the stickiness coefficient for an equivalent firm in
Switzerland. For comparison, the average degree of cost stickiness in the full sample
is -0.090, i.e., the degree of cross-country variation in cost stickiness driven by EPL has
the same order of magnitude as the average cost stickiness coefficient.
As a first robustness check, we split the full sample period (1990-2008) into two
shorter time periods, 1990-2000 and 2001-2008,30 and re-estimate Model A separately for
each subsample (columns (b) and (c) in Table 4). The estimates of β11 are negative and
significant at the 5 percent level in both subsamples (β11=-0.046, t=-2.14 between 1990-
30 The end of 2000 is a natural break point for sub-period analysis, since it is close to several major economic events, such as the introduction of the Euro at the wholesale level in early 1999, the bursting of the dot-com bubble in 2000, the collapse of Enron and related scandals that significantly increased the scrutiny on corporate governance starting in 2001. Another important factor over the same period was the Y2K concerns, which led companies to upgrade or replace their IT systems. In addition to addressing the Y2K bug, this large investment in IT allowed companies to interact in new ways with their customers and suppliers, redefining business processes and relationships (e.g., Anderson et al., 2006), which also affected cost behavior.
24
2000, and β11=-0.056, t=-2.34 between 2001-2008), lending additional support to
Hypothesis 1. In another robustness check, we re-estimate the model after discarding data
for the US firms. US data accounts for 35 percent of observations in the full sample, and
therefore it could have a disproportionate effect on the estimates. The estimation results
after discarding the US are in column (d) of Table 4. The point estimates and their
significance level (β11=-0.048, t=-4.15) are very similar to our original estimates for the
full sample. In another robustness check, we add control for other country-level labor
market characteristics (trade union density, an index of collective bargaining coordination
and centralization, and an index of unemployment benefits) that are likely to affect the
magnitude of adjustment costs for labor and consequently affect the degree of cost
stickiness. The estimation results are presented in Table 6. The estimates of β11 are close
in magnitude and significance levels to the original estimates in Table 4, indicating that
stricter EPL is associated with a significantly higher degree of cost stickiness even after
controlling for the main labor market characteristics in each country, lending additional
support to Hypothesis 1. Since cost stickiness may be affected by corporate governance
variables (e.g., Chen et al., 2011), we also add control for creditor rights index from
Djankov et al. (2007) and shareholder protection index from LaPorta et al. (2006) in one
specification, and an alternative shareholder protection index from Djankov et al. (2007)
in another specification. The main results (untabled) are similar, both with and without
control for the main labor market characteristics. Since cost stickiness may depend on the
long-term growth prospects of the firm, we also add control for long-term firm-level
growth rates,31 with similar results (untabled). In another robustness check, we replace
our main EPL strictness index from OECD (2004) with an alternative employment
protection index from Botero et al. (2004), with similar results (untabled). In a further
robustness check, we add country fixed effects and year effects, to control for unobserved
country-specific factors and time-varying unobserved global shocks. The main results
(untabled) are similar.
31 In one specification, we compute the growth rate for firm i as the average log-change in its deflated sales over the entire sample period. In another specification, we compute the growth rate for firm i in year t as the average log-change in its deflated sales between year t-4...t.
25
4.1. Results for regular and temporary employment protection measures (Model B)
As we discuss in subsection 3.1, our measure of overall employment protection
legislation in Model A (EPLn) combines two more detailed EPL indexes, employment
protection for regular employees (REGEPLn) and regulation of temporary forms of
employment (TEMPEPLn), which may have important differential effects.
We use Model B to examine the effects of regular and temporary EPL indexes on
cost stickiness. The main estimation results are presented in column (a) of Table 5. The
coefficients on the control variables (GDP growth, asset intensity and a dummy for
successive decreases in sales) have the expected signs and are consistent with the
findings in prior literature for the US data (e.g., Anderson et al., 2003). The main
parameters of interest are β12 (the coefficient on regular employment protection index
REGEPLn) and β13 (the coefficient on temporary employment regulation index
TEMPEPLn). The coefficient for regular EPL (β12) is negative but not significant at the
10 percent level (β12=-0.015, t=-1.54), while the coefficient for temporary EPL (β13) is
negative and significant at the 0.001 percent level (β13=-0.027, t=-6.20). These smaller
insignificant estimates for regular EPL suggest that companies are able to successfully
bypass most of the EPL restrictions on firing regular workers by relying on temporary
workers to accommodate most of the changes in their work force. 32 However, the
insignificant estimates for regular EPL could also be due to multicollinearity (the
correlation between regular and temporary EPL in our sample is 0.789). While the
estimates indicate that REGEPLn and TEMPEPLn are jointly significant and affect the
degree of cost stickiness in the direction consistent with Hypotheses 2a, 2b, the
inferences regarding the role of each EPL variable separately are less informative due to
multicollinearity. To get a better sense of the effects of independent variation in
REGEPLn and TEMPEPLn, we do the following. First, we regress TEMPEPLn on
REGEPLn, and compute the residual rTEMPEPLn in this regression. By construction, this
residual captures the portion of variation in TEMPEPLn that is orthogonal to
32 This does not mean that stricter regular EPL has no adverse effects for companies. For example, if temporary workers are less productive than regular workers, then stricter regular EPL will reduce the average productivity of companies’ work force by forcing them to rely more on temporary workers.
26
(uncorrelated with) REGEPLn.33 After that, we re-estimate Model B after replacing the
original TEMPEPLn with the residual rTEMPEPLn.34 The estimate of β12 in this modified
regression captures the effect of common variation in REGEPLn and TEMPEPLn, while
the estimate of β13 captures the effect of independent variation in TEMPEPLn. The
coefficients on both variables in this modified regression are negative and significant at
the 1 percent level (β12=-0.035, t=-3.17, β13=-0.027, t=-6.20), indicating that both the
common variation in regular and temporary EPL and the independent variation in
temporary EPL have a significant association with the degree of cost stickiness,
supporting Hypotheses 2a and 2b.
This association is also economically significant. For example, if regular and
temporary EPL indexes change from their levels in Switzerland (the country with the
lowest EPL in continental Europe) to their levels in France (a country with relatively
stringent EPL), it would increase the cost stickiness coefficient by 0.084 (in absolute
value), which is nearly the same as the average cost stickiness coefficient in our sample
(-0.090).
As a first robustness check, we split the full sample period (1990-2008) into two
shorter time periods, 1990-2000 and 2001-2008, and re-estimate Model B separately for
each subsample (columns (b) and (c) in Table 5). The estimates of β13 (the coefficient on
TEMPEPLn) are negative and significant in both subsamples (β13=-0.036, t=-2.45
between 1990-2000, and β13=-0.020, t=-2.84 between 2001-2008), while β12 is negative
and significant between 2001-2008 but not between 1990-2000 (β12=-0.041, t=-2.19
between 2001-2008, and β12=-0.008, t=-0.54 between 1990-2000). As discussed above,
since EPL indexes for regular and temporary employees are strongly positively correlated,
the interpretation of individual coefficients is complicated by multicollinearity.35 When
we redefine the EPL variables to separate common variation in regular and temporary
EPL from independent variation in temporary EPL using the procedure described above,
both coefficients are negative and significant at the 5 percent level in both subsamples
33 This is a general property of OLS regression: the regression residual is orthogonal to all the explanatory variables. 34 By construction, this regression is equivalent to the original Model B, since the new explanatory variable rTEMPEPLn is a linear combination of the original explanatory variables. 35 On the other hand, their combined effect is identified reliably even in the presence of strong multicollinearity.
27
(β12=-0.037, t=-2.02 and β13=-0.036, t=-2.45 between 1990-2000; β12=-0.056, t=-2.46
and β13=-0.020, t=-2.84 between 2001-2008), lending further support to Hypotheses 2a,
2b.
Since the US data accounts for 35 percent of the full sample and therefore could have
a disproportionate impact on the results, we also re-estimate the model after discarding
the US from the sample. The estimation results without the US are very similar in
magnitude and significance (column (d) in Table 5). In another robustness check, we add
control for other country-level labor market characteristics (trade union density, an index
of collective bargaining coordination and centralization, and an index of unemployment
benefits) that are likely to affect the magnitude of adjustment costs for labor and
consequently affect the degree of cost stickiness. The estimation results are presented in
Table 7. The estimates of β12, β13 are again very close in magnitude and significance
levels to the original estimates in Table 5. Like in the original estimates, the coefficient
on temporary EPL is negative and significant at any reasonable significance level
(β13=-0.023, t=-5.35) while the coefficient on regular EPL is negative but insignificant
(β12=-0.011, t=-1.24). As discussed above, the insignificant estimates for regular EPL
suggest that companies are able to bypass most of the EPL restrictions for regular
workers by relying on temporary workers, however this could also reflect
multicollinearity. When we redefine the EPL variables to separate between common
variation in regular and temporary EPL and independent variation in temporary EPL
using the procedure described above, both coefficients are negative and significant
(β12=-0.029, t=-2.65 and β13=-0.023, t=-5.35). Thus, both the common variation in
regular and temporary EPL and the independent variation in temporary EPL significantly
increase the degree of cost stickiness even after controlling for the main labor market
characteristics, further supporting Hypotheses 2a, 2b. Since cost stickiness may be
affected by corporate governance variables (e.g., Chen et al., 2011), we also add control
for creditor rights index from Djankov et al. (2007) and shareholder protection index
from LaPorta et al. (2006) in one specification, and an alternative shareholder protection
index from Djankov et al. (2007) in another specification. The main results (untabled) are
similar, both with and without control for the main labor market characteristics. Since
cost stickiness may depend on the long-term growth prospects of the firm, we also add
28
control for long-term firm-level growth rates,36 with similar results (untabled). In another
robustness check, we add country fixed effects and year effects. The main results
(untabled) are similar.
Summing up, the estimation results for a single aggregate measure of EPL (Model
A) indicate that stricter EPL is associated with a higher degree of cost stickiness,
supporting our main empirical hypothesis. When we decompose the overall EPL index
into two more detailed measures of EPL strictness, those for regular and temporary
employees (Model B), the estimates for both measures confirm our main findings. The
estimates for Model B also indicate that the association with cost stickiness is stronger for
temporary EPL than for regular EPL, which may suggest that companies are able to
successfully bypass many of the EPL restrictions for regular workers by relying more on
temporary workers.37 In both models, the association between EPL strictness and cost
stickiness is significant both statistically and economically, and the results are robust to
alternative specifications. These results support the central premise of the economic
theory of sticky costs, that cost behavior reflects deliberate resource-adjustment decisions
by forward-looking managers who recognize the dynamic tradeoffs associated with
adjustment costs, in the empirical context of labor resources.
5. Conclusion
In this study we investigated the relationship between employment protection
legislation (EPL) in different countries and sticky cost behavior. The basic premise in the
economic theory of sticky costs is that many costs arise as a result of deliberate resource
commitment decisions made by forward-looking managers in the presence of adjustment
costs. Adjustment costs play a central role in this theory, giving rise to dynamic patterns
of sticky cost behavior which are inconsistent with the standard textbook view of cost
36 In one specification, we compute the growth rate for firm i as the average log-change in its deflated sales over the entire sample period. In another specification, we compute the growth rate for firm i in year t as the average log-change in its deflated sales between year t-4...t. 37 Due to the high correlation between regular and temporary EPL, the estimates of their separate effects are less precise than the estimates of their combined effect. Therefore, more caution is required in interpreting their relative magnitudes.
29
behavior. A central testable implication of this theory is that the degree of cost stickiness
should be increasing in the magnitude of (downward) adjustment costs for capacity
resources. However, empirical tests of this theory have been hampered by the fact that
adjustment costs are hard to measure directly. In this study, we leveraged prior literature
in labor economics on the structural features of the labor market, and exploited cross-
country variation in the strictness of employment protection legislation for OECD
countries to test the theory of sticky costs. Cross-country variation in EPL strictness
provides a reliable exogenous source of variation in adjustment costs for labor resources,
since EPL is the primary source of firing costs for employers (e.g., Long and Siebert,
1983, Pissarides, 1999). Based on the economic theory of sticky costs, we hypothesized
that firms in countries with stricter employment protection should exhibit a greater degree
of cost stickiness.
We tested our hypotheses using a large sample of firms in 19 OECD countries. The
empirical results strongly support our hypotheses, reinforcing the notion that observed
cost behavior is driven by deliberate resource commitment decisions made by forward-
looking managers who recognize the role of adjustment costs. The relationship between
the strictness of EPL and the degree of cost stickiness is also highly economically
significant, and the estimation results are robust to alternative model specifications.
Our study is the first in the literature to explicitly consider and test the link between
economy-wide structural variables and sticky cost behavior. Prior literature on cost
behavior was conducted almost exclusively using samples of firms from a single country,
and as such has largely ignored the impact of economy-wide structural variables on cost
behavior.38 Our results show that a full understanding of cost behavior in general and of
cost stickiness in particular requires careful analysis not only of the firm-specific factors
but also of the economy-wide structural forces that shape managers’ resource adjustment
decisions.
38 The only exception we are aware of is Calleja et al. (2006).
30
References
Anderson, M.C., Banker, R.D., Janakiraman, S.N., 2003. Are selling, general, and
administrative costs "sticky"? Journal of Accounting Research. 41(1), 47-63.
Anderson, M.C., Banker, R.D., Ravindran, S., 2006. Value implications of investment in
information technology. Management Science. 52(9), 1359-1376.
Autor, D.H., Kerr, W.R., Kugler, A.D., 2007. Do employment protections reduce
productivity? Evidence from U.S. states. Economic Journal. 117, F189-F217.
Balakrishnan, R., Peterson, M., Soderstrom, N., 2004. Does capacity utilization affect the
"stickiness" of cost? Journal of Accounting, Auditing and Finance. 19(3), 283-299.
Banker, R.D., Byzalov, D., Plehn-Dujowich, J.M., 2010a. Sticky cost behavior: Theory
and Evidence. Working paper.
Banker, R.D., Chen, L., 2006. Predicting earnings using a model based on cost variability
and cost stickiness. Accounting Review. 81(2), 285-307.
Banker, R.D., Ciftci, M., Mashruwala, R., 2010b. Managerial optimism, prior period
sales changes, and sticky cost behavior. Working paper.
Barone, A., 2001. Employment protection legislation: a critical review of the literature.
Working paper.
Bassanini, A., Nunziata, L., Venn, D., 2009. Job protection legislation and productivity
growth in OECD countries. Economic Policy. 24, 349-402.
Bentolila, S., Bertola, G., 1990. Firing costs and labor demand: How bad is
Eurosclerosis? Review of Economic Studies. 57, 381-402.
Besley, T., Burgess, R., 2004. Can Labor Regulation Hinder Economic Performance?
Evidence from India. Quarterly Journal of Economics. 119(1), 91-134.
Blanchard, O., Portugal, P., 2001. What hides behind an unemployment rate: Comparing
Portuguese and U.S. labor markets. American Economic Review. 91(1), 187-207.
Boeri, T., Jimeno, J., 2005. The effects of employment protection: learning from variable
enforcement. European Economic Review. 49, 2057-77.
31
Botero, J., Djankov, S., La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2004. The
Regulation of Labor. Quarterly Journal of Economics. 119(4), 1339-1382.
Burgess, S., Knetter, M., Michelacci, C., 2000. Employment and output adjustment in the
OECD: a disaggregate analysis of the role of job security provisions. Economica. 67,
419-35.
Caballero, R., Cowan, K., Engel, E., Micco, A., 2004. Effective labor regulation and
microeconomic flexibility. Cowles Foundation Discussion Papers 1480.
Calleja, K., Steliaros, M., Thomas, D.C., 2006. A note on cost stickiness: some
international comparisons. Management Accounting Research. 17, 127-140.
Chen, C., Lu, H., Sougiannis, T., 2011. The agency problem, corporate governance, and
the asymmetrical behavior of selling, general and administrative costs. Contemporary
Accounting Research, forthcoming.
Cingano, F., Leonardi, M., Messina, J., Pica, G., 2008. Employment protection legislation,
productivity and investment: evidence from Italy. Working paper.
DeFreitas, G., Marshall, A., 1998. Labor surplus, worker rights and productivity growth:
a comparative analysis of Asia and Latin America. Labor. 12(3), 515-39.
Dierynck, B., Renders, A., 2009. Earnings Management Incentives and the Asymmetric
Behavior of Labor Costs in Belgium. Working paper.
Djankov, S., McLiesh, C., Shleifer, A., 2007. Private credit in 129 countries. Journal of
Financial Economics. 84, 299-329.
Goux, D., Maurin, E., Pauchet, M., 2001, Fixed-term contracts and the dynamics of labor
demand. European Economic Review. 45, 533-552.
Haltiwanger, J., Scarpetta, S., Schweiger, H., 2006. Assessing job flows across countries:
the role of industry, firm-size and regulations. IZA Discussion Paper No. 2450.
Hamermesh, D., Pfann, G.A., 1996. Adjustment costs in factor demand. Journal of
Economic Literature. 34, 1264-1292.
Heckman, J., Pagés-Serra, C., Edwards, A., Guidotti, P., 2000. The Cost of Job Security
Regulation: Evidence from Latin American Labor Markets. Economia. 1(1), 109-154.
32
Hopenhayn, H., Rogerson, R., 1993. Job turnover and policy evaluation: a general
equilibrium analysis. Journal of Political Economy. 101(5), 915-38.
Kama, I., Weiss, D., 2010. Do managers' deliberate decisions induce sticky costs?
Working paper.
Kugler, A., Pica, G., 2008. Effects of employment protection on worker and job flows:
evidence from the 1990 Italian reform. Labor Economics. 15(1), 78-95.
LaPorta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R., 1997. Legal determinants of
external finance. Journal of Finance. 52(3), 1131-1150.
LaPorta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R., 1998. Law and finance.
Journal of Political Economy. 106(6), 1113-1155.
LaPorta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R., 2000. Investor protection and
corporate governance. Journal of Financial Economics. 58, 3-27.
LaPorta, R., Lopez-de-Silanes, F., Shleifer, A., 2006. What works in securities laws?
Journal of Finance. 61, 1-32.
Lazear, E., 1990. Job security provisions and employment. Quarterly Journal of
Economics. 105(3), 699-726.
Long, V.N., Siebert, H., 1983. Layoff restraints and the demand for labor. Zeitschrift fur
die gesamte Staatswissenschaft. 139(4), 612-624.
Medeiros, O.R., Costa, P.S., 2005. Cost stickiness in Brazilian firms. Working paper.
Micco, A., Pages, C., 2006. The economic effects of employment protection: evidence
from international industry-level data. IZA Discussion Paper No. 2433.
Mortensen, D.T., Pissarides, C.A., 1999. Unemployment Responses to ‘Skill-biased’
Technology Shocks: the Role of Labor Market Policy. The Economic Journal. 109,
242-265.
Nickell, S., 1997. Unemployment and labor market rigidities: Europe versus North
America. Journal of Economic Perspectives. 11(3), 55-74.
33
Nickell, S., Nunziata, L., Ochel, W., 2005. Unemployment in the OECD since the 1960s:
What do we know? The Economic Journal. 115, 1-27.
Noreen, E., Soderstrom, N., 1997. The accuracy of proportional cost models: Evidence
from hospital service departments. Review of Accounting Studies. 2, 89-114.
OECD, 2004. Employment outlook. Paris: OECD.
Pierre, G., Scarpetta, S., 2006. Employment protection: do firms’ perceptions match with
legislation? Economics Letters. 90, 328-34.
Pissarides, C.A., 1999. Policy influences on unemployment: The European experience.
Scottish Journal of Political Economy. 46(4), 389-418.
Rogers, W.H., 1993. Regression standard errors in clustered samples. Stata Technical
Bulletin. 13, 19-23.
Schivardi, F., Torrini, R., 2008. Identifying the effects of firing restrictions through size-
contingent differences in regulation. Labor Economics. 15(3), 482-511.
Weiss, D., 2010. Cost behavior and analysts' earnings forecasts. The Accounting Review.
85(4), 1441-1471.
34
Table 1. Calculation of the summary indicators of EPL strictness reported in OECD (2004)
For each country, EPL is characterized along 14 basic items described in Panels A and B below. As a first step in computing these indexes in OECD (2004), each basic item is renormalized into a
cardinal score ranging from 0 to 6, where higher values represent stricter regulation. After that, the summary indices of EPL are computed as a weighted average of individual items, with weights described in Panel C below.
Panel A. EPL for regular employees (source: Table 2.A1.1 in OECD 2004) assignment of numerical scores of strictness short description 0 1 2 3 4 5 6
Item 1 Notification procedures
Scale 0-3 0 when an oral statement is enough; 1 when a written statement of the reasons for dismissal must be supplied to the employee; 2 when a third party (such as works council or the competent labor authority) must be notified; 3 when the employer cannot proceed to dismissal without authorization from a third party.
Scale (0 – 3) × 2
Item 2 Delay involved before notice can start
Days Estimated time includes, where relevant, the following assumptions: 6 days are counted in case of required warning procedure, 1 day when dismissal can be notified orally or the notice can be directly handed to the employee, 2 days when a letter needs to be sent by mail and 3 days when this must be a registered letter.
≤2 <10 <18 <26 < 35 <45 ≥45
Item 3 Length of the notice period at
9 months tenure Months 4 years tenure Months 20 years tenure Months
0 0
<1
≤0.4 ≤0.75 ≤2.75
≤0.8 ≤1.25
<5
≤1.2 <2 <7
<1.6 <2.5 <9
<2 <3.5 <11
≥2 ≥3.5 ≥11
Item 4 Severance pay at
9 months tenure Months pay 4 years tenure Months pay 20 years tenure Months pay
0 0 0
≤0.5 ≤0.5 ≤3
≤1 ≤1 ≤6
≤1.75 ≤2 ≤10
≤2.5 ≤3 ≤12
<3 <4 ≤18
≥3 ≥4
>18 Item 5 Definition of justified or unfair dismissal
Scale 0-3 0 when worker capability or redundancy of the job are adequate and sufficient ground for dismissal; 1 when social considerations, age or job tenure must when possible influence the choice of which worker(s) to dismiss; 2 when a transfer and/or a retraining to adapt the worker to different work must be attempted prior to dismissal; 3 when worker capability cannot be a ground for dismissal.
Scale (0 – 3) × 2
Item 6 Length of trial period
Months Period within which, regular contracts are not fully covered by employment protection provisions and unfair dismissal claims can usually not be made.
≥24 >12 >9 >5 >2.5 ≥1.5 <1.5
Item 7 Compensation following unfair dismissal
Months pay ≤3 ≤8 ≤12 ≤18 ≤24 ≤30 >30
Item 8 Possibility of reinstatement following unfair dismissal
Scale 0-3 The extend of reinstatement is based upon whether, after finding of unfair dismissal, the employee has the option of reinstatement into his/her previous job, even if this is against the wishes of the employer.
Scale (0 – 3) × 2
35
Panel B. EPL for temporary employees (source: Table 2.A1.1 in OECD 2004) assignment of numerical scores of strictness short description 0 1 2 3 4 5 6
Item 9 Valid cases for use of fixed-term contracts (FTC)
Scale 0-3 0 fixed-term contracts are permitted only for “objective” or “material situation”, i.e. to perform a task which itself is of fixed duration; 1 if specific exemptions apply to situations of employer need (e.g. launching a new activity) or employee need (e.g. workers in search of their first job); 2 when exemption exist on both the employer and employee sides; 3 when there are no restrictions on the use of fixed-term contracts.
6 – scale (0 – 3) × 2
Item 10 Maximum number of successive FTC
Number no limit
≥5 ≥4 ≥3 ≥2 ≥1.5 <1.5
Item 11 Maximum cumulated duration of successive FTC
Months no limit
≥36 ≥30 ≥24 ≥18 ≥12 <12
Item 12 Types of work for which temporary work agency (TWA) employment is legal
Scale 0-4 0 when TWA employment is illegal; 1-3 1 to 3 depending upon the degree of restrictions; 4 when no restrictions apply. 6 – Scale (0 – 4) × 6/4
Item 13 Restrictions on number of renewals
Yes/no - - No - Yes - -
Item 14 Maximum cumulated duration of TWA contracts
Months no limit
≥36 ≥24 ≥18 ≥12 ≥6 <6
36
Panel C. EPL summary indicators at different levels of aggregation and the weighting scheme
(source: Table 2.A1.2 in OECD 2004)
Level 4 Scale 0-6
Level 3 Scale 0-6
Level 2 Scale 0-6
Level 1 Scale 0-6
Procedural inconveniences (1/3)
1. Notification procedures (1/2) 2. Delay to start a notice (1/2)
Notice and severance pay for no-fault individual dismissals (1/3)
3. Notice period after 9 months (1/7) 4 years (1/7) 20 years (1/7) 4. Severance pay after 9 months (4/21) 4 years (4/21) 20 years (4/21)
Regular contracts REGEPLn (1/2)
Difficulty of dismissal (1/3)
5. Definition of unfair dismissal (1/4) 6. Trial period (1/4) 7. Compensation (1/4) 8. Reinstatement (1/4)
Fixed term contracts (1/2) 9. Valid cases for use of fixed-term contracts (1/2) 10. Maximum number of successive contracts (1/4) 11. Maximum cumulated duration (1/4)
Overall summary indicator EPLn
Temporary contracts TEMPEPLn (1/2)
Temporary work agency employment (1/2)
12. Types of work for which is legal (1/2) 13. Restrictions on number of renewals (1/4) 14. Maximum cumulated duration (1/4)
37
Table 2. Variable definitions
n – country index. i – firm index. t – year index. XOPRn,i,t – operating costs for firm i in country n in year t, deflated to control for inflation. SALEn,i,t – sales revenue for firm i in country n in year t, deflated to control for inflation. GDPGROWTHn,t – real GDP growth in country n in year t. DECn,i,t – binary variable equal to one for sales decreases. AINTn,i,t – asset intensity for firm i in country n in year t, computed as the log ratio of assets to sales,
ln(AT/SALE). LAWn – binary variable equal to one for common-law countries (Australia, Canada, Ireland, UK and US),
zero otherwise. REGEPLn – index of employment protection legislation (EPL) for regular employees in country n in late
1990-s, from Table 2.A2.4 in OECD (2004). REGEPLn ranges from 0 to 6, and higher values correspond to stricter EPL.
TEMPEPLn – index of employment protection legislation (EPL) for temporary employees in country n in late 1990-s, from Table 2.A2.4 in OECD (2004). TEMPEPLn ranges from 0 to 6, and higher values correspond to stricter EPL.
EPLn – aggregate index of employment protection legislation (EPL) in country n in late 1990s, computed as (REGEPLn+TEMPEPLn)/2 following OECD (2004).
TUDn – trade union density in country n in 2000, from OECD (2004). BCCn – bargaining coordination and centralization index for country n in late 1990s, from OECD (2004). BNFTn – unemployment benefits index for country n in late 1990-s, from Nickell et al. (2005).
Sample selection criteria: We start by drawing the full sample of publicly-listed non-financial firms from Compustat (Global
and North America) between 1988-2008 for the 19 OECD countries we focus on. We discard firm-years if: (1) sales or operating costs are missing or negative in current or two prior years, (2) operating costs are less than 50% or more than 200% of sales in current or two prior years, or (3) assets are missing or negative in current year. We also discard firms reporting in a non-native currency (for example, European firms reporting in US dollars). After that, we discard 1% outliers on each tail for the dependent variable (log-change in deflated operating costs) and for the continuous firm-level explanatory variables (log-change in deflated sales and asset intensity). We also discard firm-years if deflated sales increased by over 50% or dropped by over 33% in current or prior year, since such extreme year-on-year changes in sales likely reflect mergers or divestitures. The final sample in estimation contains 128,333 observations for 15,833 firms in 19 OECD countries between 1990-2008.42
42 The first two lags in the original data is used up to compute first differences and control variables, so the final sample in estimation starts in 1990 rather than in 1988.
38
Table 3. Descriptive statistics
country number of observations
average ∆lnCOST
average ∆lnSALE
average GDP growth
regular EPL (REGEPL)
temporary EPL
(TEMPEPL) Australia 3,798 0.045 0.041 3.5 1.5 0.9 Austria 755 0.051 0.048 2.4 2.9 1.5 Belgium 894 0.031 0.029 2.1 1.7 2.6 Canada 4,651 0.046 0.042 2.7 1.3 0.3 Denmark 1,254 0.038 0.036 2.0 1.5 1.4 Finland 1,192 0.050 0.049 2.9 2.3 1.9 France 5,941 0.042 0.040 1.9 2.3 3.6 Germany 5,613 0.030 0.029 1.6 2.7 2.3 Ireland 478 0.053 0.053 6.2 1.6 0.3 Italy 1,743 0.035 0.031 1.1 1.8 3.6 Japan 37,094 0.022 0.022 1.3 2.4 1.6 Netherlands 1,516 0.035 0.035 2.7 3.1 1.2 Norway 960 0.040 0.042 2.8 2.3 3.1 Portugal 414 0.027 0.029 1.9 4.3 3 Spain 1,187 0.043 0.042 3.2 2.6 3.3 Sweden 2,096 0.050 0.051 2.5 2.9 1.6 Switzerland 1,898 0.031 0.030 1.7 1.2 1.1 UK 12,286 0.035 0.032 2.5 0.9 0.3 US 44,563 0.041 0.039 2.8 0.2 0.3
The variable definitions are described in Table 2.
39
Table 4. Estimates of the relationship between the aggregate EPL index and cost stickiness
The estimation model is
tintintinn
ntintntin
tinnntintntin
SALEDECEPLLAWAINTGDPGROWTHDEC
SALEEPLLAWAINTGDPGROWTHXOPR
,,,,,,11
10,,9,81,,76
,,54,,3,210,,
ln)(
ln)(ln
where ∆lnXOPRn,i,t is the log-change in deflated operating costs for firm i in country n in year t, ∆lnSALEn,i,t is the log-change in deflated sales, GDPGROWTHn,t is the real GDP growth rate in country n in year t, AINTn,i,t is asset intensity (log ratio of assets to sales), LAWn is a binary variable equal to one for common-law countries, EPLn is the aggregate employment protection legislation index for country n, DECn,i,t is a binary variable equal to one if deflated sales decreased in year t, and εn,i,t is an error term. The sample selection criteria are described in Table 2.
main sample 1990-2008
subsample 1990-2000
subsample 2001-2008
main sample excluding US exp.
sign (a) (b) (c) (d)
β0 0.001 (0.92)
0.000 (0.07)
0.002*** (3.13)
0.000 (-0.30)
β1 + 0.913*** (62.68)
0.887*** (29.86)
0.913*** (56.90)
0.935*** (106.22)
β2 + 0.004** (2.00)
0.005 (1.12)
-0.010* (-1.83)
0.005 (1.56)
β3 – -0.033*** (-4.01)
-0.028*** (-2.72)
-0.033*** (-4.14)
-0.032** (-1.99)
β4 0.021 (0.83)
0.095* (1.87)
-0.013 (-0.54)
0.007 (0.27)
β5 0.019* (1.65)
0.049** (2.34)
0.010 (0.82)
0.004 (0.28)
β6 – -0.081*** (-6.21)
-0.057** (-2.35)
-0.057** (-2.08)
-0.103*** (-11.24)
β7 + 0.134*** (7.74)
0.124*** (3.58)
0.134*** (12.02)
0.111*** (10.32)
β8 – -0.001 (-0.28)
-0.008 (-1.47)
0.025*** (3.46)
0.001 (0.35)
β9 – -0.080*** (-6.36)
-0.091*** (-5.09)
-0.079*** (-5.76)
-0.075*** (-3.51)
β10 -0.125*** (-5.22)
-0.175*** (-4.10)
-0.123** (-2.52)
-0.127*** (-5.42)
β11 (EPL) – -0.044*** (-4.03)
-0.046** (-2.14)
-0.056** (-2.34)
-0.048*** (-4.15)
N 128,333 63,751 64,582 83,649 adj. R2 0.7925 0.8144 0.7722 0.7964 t-values in parentheses. The t-values are computed using clustered standard errors (Rogers 1993) with
clustering by country. *, ** and *** indicate significance at 10, 5 and 1 percent levels respectively.
40
Table 5. Estimates of the relationship between EPL indexes for regular and temporary employees and cost stickiness
The estimation model is
tintintinnn
ntintntin
tinnnntintntin
SALEDECTEMPEPLREGEPLLAWAINTGDPGROWTHDEC
SALETEMPEPLREGEPLLAWAINTGDPGROWTHXOPR
,,,,,,1312
11,,10,91,,87
,,654,,3,210,,
ln)(
ln)(ln
where ∆lnXOPRn,i,t is the log-change in deflated operating costs for firm i in country n in year t, ∆lnSALEn,i,t is the log-change in deflated sales, GDPGROWTHn,t is the real GDP growth rate in country n in year t, AINTn,i,t is asset intensity (log ratio of assets to sales), LAWn is a binary variable equal to one for common-law countries, REGEPLn is the employment protection legislation index for regular employees for country n, TEMPEPLn is a measure of temporary employment regulation in country n, DECn,i,t is a binary variable equal to one if deflated sales decreased in year t, and εn,i,t is an error term. The sample selection criteria are described in Table 2.
main sample 1990-2008
subsample 1990-2000
subsample 2001-2008
main sample excluding US exp.
sign (a) (b) (c) (d)
β0 0.001 (0.94)
0.000 (0.08)
0.002*** (3.14)
0.000 (-0.27)
β1 + 0.914*** (60.35)
0.887*** (30.24)
0.912*** (54.08)
0.940*** (186.56)
β2 + 0.004** (2.01)
0.005 (1.12)
-0.010* (-1.81)
0.006** (2.02)
β3 – -0.033*** (-3.98)
-0.028*** (-2.68)
-0.033*** (-4.10)
-0.033** (-2.07)
β4 0.019 (0.71)
0.094* (1.83)
-0.011 (-0.44)
-0.018 (-0.91)
β5 0.006 (0.59)
0.022 (1.29)
0.007 (0.59)
-0.028*** (-2.76)
β6 0.012* (1.95)
0.027** (2.13)
0.004 (0.50)
0.011* (1.88)
β7 – -0.083*** (-5.74)
-0.059*** (-2.59)
-0.051* (-1.85)
-0.104*** (-10.63)
β8 + 0.134*** (7.77)
0.123*** (3.58)
0.134*** (11.78)
0.111*** (10.30)
β9 – -0.001 (-0.28)
-0.007 (-1.36)
0.026*** (3.40)
0.000 (0.09)
β10 – -0.079*** (-6.23)
-0.090*** (-5.01)
-0.080*** (-5.66)
-0.074*** (-3.45)
β11 -0.122*** (-4.59)
-0.169*** (-4.14)
-0.133*** (-2.65)
-0.119*** (-4.34)
β12 (REGEPL) – -0.015 (-1.54)
-0.008 (-0.54)
-0.041** (-2.19)
-0.016 (-1.13)
β13 (TEMPEPL) – -0.027*** (-6.20)
-0.036** (-2.45)
-0.020*** (-2.84)
-0.026*** (-6.21)
N 128,333 63,751 64,582 83,649 adj. R2 0.7925 0.8144 0.7722 0.7966
t-values in parentheses. The t-values are computed using clustered standard errors (Rogers 1993) with clustering by country. *, ** and *** indicate significance at 10, 5 and 1 percent levels respectively.
41
Table 6. Estimates of the relationship between the aggregate EPL index and cost stickiness, after controlling for other labor market characteristics
The estimation model is
tintintinnn
nnntintntin
tinnn
nnntintntin
SALEDECEPLBNFTBCCTUDLAWAINTGDPGROWTHDEC
SALEEPLBNFTBCCTUDLAWAINTGDPGROWTHXOPR
,,,,,,1716
151413,,12,111,,109
,,87
654,,3,210,,
ln)(
ln)(ln
where ∆lnXOPRn,i,t is the log-change in deflated operating costs for firm i in country n in year t, ∆lnSALEn,i,t is the log-change in deflated sales, GDPGROWTHn,t is the real GDP growth rate in country n in year t, AINTn,i,t is asset intensity (log ratio of assets to sales), LAWn is a binary variable equal to one for common-law countries, TUDn is trade union density in country n, BCCn is the bargaining centralization and coordination index for country n, BNFTn is the unemployment benefits index for country n, EPLn is the aggregate employment protection legislation index for country n, DECn,i,t is a binary variable equal to one if deflated sales decreased in year t, and εn,i,t is an error term. The sample selection criteria are described in Table 2.
main sample 1990-2008
subsample 1990-2000
subsample 2001-2008
main sample excluding US exp.
sign (a) (b) (c) (d)
β0 0.001 (1.00)
0.000 (0.17)
0.002*** (3.01)
0.000 (-0.18)
β1 + 0.912*** (54.14)
0.885*** (40.19)
0.891*** (34.98)
0.939*** (60.96)
β2 + 0.004** (1.96)
0.005 (1.36)
-0.011** (-2.00)
0.005 (1.55)
β3 – -0.033*** (-3.94)
-0.026*** (-2.73)
-0.033*** (-4.32)
-0.032** (-1.97)
β4 0.009 (0.39)
0.060* (1.94)
0.004 (0.16)
-0.010 (-0.39)
β5 0.025 (0.55)
0.132** (2.49)
-0.045 (-1.22)
-0.010 (-0.33)
β6 -0.007 (-1.20)
-0.026*** (-3.18)
0.014** (1.96)
-0.008 (-1.24)
β7 0.029 (1.55)
0.068* (1.79)
-0.002 (-0.08)
0.035** (2.23)
β8 0.013 (1.18)
0.033** (2.17)
0.011 (0.81)
-0.004 (-0.36)
β9 – -0.062*** (-2.96)
-0.063** (-2.05)
0.017 (0.40)
-0.074*** (-3.83)
β10 + 0.134*** (7.75)
0.123*** (3.56)
0.134*** (11.63)
0.111*** (9.96)
β11 – 0.000 (-0.05)
-0.007 (-1.43)
0.029*** (3.90)
0.003 (0.55)
β12 – -0.080*** (-6.48)
-0.093*** (-5.67)
-0.079*** (-5.96)
-0.075*** (-3.58)
β13 -0.121*** (-4.78)
-0.132*** (-3.68)
-0.171*** (-3.58)
-0.129*** (-5.10)
β14 -0.006 (-0.17)
-0.064 (-1.06)
0.008 (0.28)
-0.035 (-1.48)
β15 0.001 (0.17)
0.030*** (2.80)
-0.029*** (-2.89)
0.000 (0.06)
β16 -0.043*** (-2.59)
-0.107*** (-3.81)
0.008 (0.30)
-0.032 (-1.60)
β17 (EPL) – -0.036*** (-3.26)
-0.028 (-1.50)
-0.060*** (-3.19)
-0.046*** (-4.42)
N 128,333 63,751 64,582 83,649 adj. R2 0.7926 0.8147 0.7723 0.7965 t-values in parentheses. The t-values are computed using clustered standard errors (Rogers 1993) with
clustering by country. *, ** and *** indicate significance at 10, 5 and 1 percent levels respectively.
42
Table 7. Estimates of the relationship between EPL indexes for regular and temporary employees and cost stickiness, after controlling for other labor market characteristics
The estimation model is
tintintinnnn
nnntintntin
tinnnn
nnntintntin
SALEDECTEMPEPLREGEPLBNFTBCCTUDLAWAINTGDPGROWTHDEC
SALETEMPEPLREGEPLBNFTBCCTUDLAWAINTGDPGROWTHXOPR
,,,,,,191817
161514,,13,121,,1110
,,987
654,,3,210,,
ln)(
ln)(ln
where ∆lnXOPRn,i,t is the log-change in deflated operating costs for firm i in country n in year t, ∆lnSALEn,i,t is the log-change in deflated sales, GDPGROWTHn,t is the real GDP growth rate in country n in year t, AINTn,i,t is asset intensity (log ratio of assets to sales), LAWn is a binary variable equal to one for common-law countries, TUDn is trade union density in country n, BCCn is the bargaining centralization and coordination index for country n, BNFTn is the unemployment benefits index for country n, REGEPLn is the employment protection legislation index for regular employees for country n, TEMPEPLn is the employment protection regulation index for temporary employees in country n, DECn,i,t is a binary variable equal to one if deflated sales decreased in year t, and εn,i,t is an error term. The sample selection criteria are described in Table 2.
main sample 1990-2008
subsample 1990-2000
subsample 2001-2008
main sample excluding US exp.
sign (a) (b) (c) (d) β0 0.001
(1.01) 0.000 (0.20)
0.002*** (3.02)
0.000 (-0.15)
β1 + 0.912*** (52.16)
0.883*** (40.55)
0.889*** (34.33)
0.945*** (58.78)
β2 + 0.004** (1.99)
0.005 (1.36)
-0.011** (-1.99)
0.006* (1.91)
β3 – -0.033*** (-3.96)
-0.027*** (-2.75)
-0.033*** (-4.24)
-0.033** (-2.03)
β4 0.007 (0.32)
0.058* (1.82)
0.007 (0.25)
-0.029 (-1.28)
β5 0.034 (0.74)
0.156*** (2.72)
-0.055 (-1.25)
0.001 (0.05)
β6 -0.007 (-1.18)
-0.025*** (-3.01)
0.014** (2.02)
-0.007 (-1.15)
β7 0.025 (1.30)
0.060 (1.48)
0.003 (0.08)
0.023** (2.10)
β8 0.002 (0.28)
0.006 (0.41)
0.010 (0.86)
-0.028*** (-2.89)
β9 0.011 (1.35)
0.027** (2.33)
0.001 (0.12)
0.007 (1.14)
β10 – -0.063*** (-2.70)
-0.062* (-1.95)
0.021 (0.50)
-0.074*** (-3.46)
β11 + 0.134*** (7.76)
0.123*** (3.56)
0.134*** (11.51)
0.111*** (10.01)
β12 – 0.000 (-0.06)
-0.007 (-1.39)
0.030*** (3.86)
0.002 (0.37)
β13 – -0.079*** (-6.39)
-0.093*** (-5.56)
-0.080*** (-5.84)
-0.075*** (-3.53)
β14 -0.118*** (-4.18)
-0.126*** (-3.34)
-0.177*** (-3.91)
-0.124*** (-4.28)
β15 -0.019 (-0.58)
-0.102* (-1.81)
0.029 (0.70)
-0.041* (-1.73)
β16 0.001 (0.16)
0.030*** (3.16)
-0.029*** (-3.08)
0.000 (0.08)
β17 -0.037** (-2.30)
-0.094*** (-3.22)
-0.003 (-0.09)
-0.028 (-1.30)
β18 (REGEPL) – -0.011 (-1.24)
0.004 (0.33)
-0.041*** (-2.85)
-0.017 (-1.45)
β19 (TEMPEPL) – -0.023*** (-5.35)
-0.029*** (-2.64)
-0.021** (-2.08)
-0.025*** (-4.58)
N 128,333 63,751 64,582 83,649 adj. R2 0.7926 0.8147 0.7723 0.7966
t-values in parentheses. The t-values are computed using clustered standard errors (Rogers 1993) with clustering by country. *, ** and *** indicate significance at 10, 5 and 1 percent levels respectively.