Review of Economic Dynamics 9 (2006) 224–241

www.elsevier.com/locate/red

Employment protection and high-tech aversion

Roberto M. Samaniego
George Washington University, Department of Economics, 1922 F Street NW, Suite 208, Washington, DC 20052
Received 3 June 2004; revised 6 October 2005
Available online 28 November 2005

Abstract
Do institutional firing costs slow the diffusion of information and communications technology (ICT)?
The paper develops a model in which, as the technology at a given plant drops behind the best practice, it
optimally reduces its workforce. As a result, firing costs are particularly detrimental to profits in industries
in which the rate of technical change is rapid—such as ICT—and countries with high firing costs specialize
in industries in which technical change is sluggish. The paper suggests that industry composition is a new
channel through which labor market regulation might impact macroeconomic aggregates.
 2005 Elsevier Inc. All rights reserved.

JEL classification: J32; J65; L16; L63; O33; O38

Keywords: Employment protection; Firing costs; Rate of technical change; Information and communications
technology; Industry composition

1. Introduction

It is widely known that the prevalence of information and communications technologies (ICT)
varies across countries. This phenomenon can be observed not only between “developed” and
“developing” economies, but across industrialized countries also. For example, Pilat and Lee
(2001) report that, among OECD economies, the number of personal computers (PCs) per 100
inhabitants in 1999 ranged from 65 in the United States down to 10 in Spain and Portugal.
This paper argues that a factor behind these differences in ICT diffusion could be differ-
ences across countries in employment protection legislation (EPL). Countries vary substantially
in terms of EPL and, as discussed below, there is a strong correlation between the presence of
these policies and the slow diffusion of ICT.

E-mail address: roberto@gwu.edu.

1094-2025/$ – see front matter  2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.red.2005.10.002
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 225

The theory developed in this paper works as follows. Suppose that industries differ among
each other in terms of the rate of technical change. In an industry in which technical change is
rapid, plants will tend to fall behind the frontier technology faster than in industries in which
technical change is slow. As a consequence, the optimal plant size will decline at a pace that is
linked to the rate of technical change—indeed, Mitchell (2002) finds that cross-industry US data
is consistent with such a link. This implies that a given job will be destroyed sooner in industries
in which technical change is rapid: the effects of firing costs imposed by EPL should then be most
severe in such industries. Several studies show that the rate of technical change does in fact vary
significantly across industries—see Gordon (1990), Bartelsman and Gray (1996), Cummins and
Violante (2002), and Wilson (2002). In particular, this rate appears to be most rapid in industries
that employ ICT intensively—such as communications, computers and electronics. If the above
intuition regarding the relationship between the rate of technical change and EPL is correct, it
implies that, ceteris paribus, ICT should be less prevalent in countries in which dismissal costs
are high.
The paper has two main contributions. First, although an extensive literature addresses the
impact of EPL upon labor markets and aggregate income, the potential cross-industry effects of
EPL have not been addressed. Second, the broader implication of the results is that the equilib-
rium industry composition constitutes a new and potentially important channel through which
EPL and possibly other forms of regulation might affect macroeconomic aggregates.
Oliner and Sichel (2000) attribute a large part of the resurgence in US economic growth in the
late 1990s to the diffusion of ICT, while Colecchia and Schreyer (2002) find that this phenom-
enon does not extend to all industrialized countries. This suggests that the observed differences
in ICT diffusion may have significant macroeconomic consequences.1 This paper finds that em-
ployment protection policies could be a factor behind these differences.
I conclude the introduction with some evidence of a link between EPL and the slow diffusion
of ICT. Nicoletti et al. (2000) construct an index of EPL. The main components of the index are
mandated severance pay and advance notice requirements, each of which can be shown to act as
firing costs under simple assumptions—whence the attention devoted to them in the literature.2
Results are reported for four indicators of ICT diffusion. Some direct indicators are available,
such as the number of personal computers per capita, and the share of ICT in aggregate spending.
In addition, one use of ICT capital for which there are no direct substitutes is e-commerce: hence,
the paper also uses the log number of internet hosts and the log number of secure servers relative
to the population as measures of the prevalence of e-commerce infrastructure.3
Figure 1 reveals a striking negative relationship between EPL and measures of ICT diffusion.
The correlations are all negative and significant, ranging from −60% for PCs and hosts to about

1 There is an independent literature that addresses differences in access to and usage of ICT across households within a
country, and the influence of ICT upon markets for different types of labor. On this, see for example Autor et al. (2003).
2 See Addison and Teixeira (2003) for a survey. Also, see Gust and Marquez (2004) for further evidence of the link
between EPL and ICT.
3 An internet host is any computer with full two-way access to the network, whereas a secure server is any computer
that contains websites that may be accessed over the internet and which supports encryption.
Data are available for 20 OECD countries. An advantage of concentrating on industrialized economies is that it is
not unreasonable to assume that they can draw from a similar set of technologies. The included countries are: Australia
(AUS), Austria (AUT), Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Netherlands, New
Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the UK and the US. Sources: Coppel (2000) and Pilat and Lee
(2001).
226 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

Fig. 1. Employment protection and information and communications technology.

−80% for secure servers and ICT spending.4 Spain, Italy and Portugal have the highest levels of
EPL and the lowest diffusion of ICT, whereas the US has the lowest level of EPL and the highest
diffusion of ICT by most measures.
Section 2 introduces dismissal costs into an industry model in which the optimal scale of
production and the rate of technical change are related. Section 3 shows how this relationship
can imply cross-country differences in industry composition. Section 4 extends the model to
allow for learning over the life cycle, and Section 5 provides an illustration of the quantitative
implications of the results. Section 6 concludes.

4 The channel upon which I focus involves the rate of capital-embodied technical change: if this is truly the parameter
of interest, then where the capital is produced should not matter, merely where it is used. Consistent with this hypothesis,
Colecchia and Schreyer (2002) do not find that a large domestic ICT-producing sector is linked to an important aggregate
role for ICT. Similarly, the correlation between EPL and the share of ICT in private sector employment (as reported in
Pilat and Lee, 2001) is a negative but weak −16%.
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 227

2. Theoretical model

This section introduces EPL in the form of dismissal costs into a simple vintage capital model.
The key feature of the model is that a plant’s optimal size declines over its lifetime at a pace that
is tied to the rate of technical change. This is because the productivity of a plant decreases after its
inception with respect to that of the frontier technology, consistent with the evidence of Mitchell
(2002).

2.1. Production and EPL

Time is continuous. Establishments in a given industry are endowed with the following pro-
duction function:
yτ = egv nατ , (1)
where τ is the date, yτ is output, nτ is employment, and v is the vintage of its technology, the
date at which the establishment was created. The key parameter is g, the rate of technical change.
The value of g is assumed to be industry-specific and not to change over time.
The plant pays a wage w, and charges a price pτ for its output at any date τ , which it takes as
given.5 pτ , w and g are positive numbers and 0 < α < 1.
There is a linear job destruction cost κw > 0. For instance, κ could be mandated severance
pay (relative to the wage) to which dismissed workers are entitled by law. The dismissal cost κ
will be the index of EPL intensity in a given country.
Plants discount the future at rate ρ, and die in accordance with a Poisson process with para-
meter ζ . The paper assumes that the dismissal cost need not be paid upon exit.6
Define r ≡ ρ + ζ : the following assumption is necessary for there to be a solution in which
hiring any worker at all is profitable.

Assumption 1. κr < 1.

We also assume henceforth that pτ = p0 e−gτ . This implies a stationary industry environment.
Mitchell (2002) shows in a related context that this is consistent with demand being unit-elastic.
Finally let t ≡ τ − v be the plant’s age. This will simplify the notation in what follows.
In the absence of firing costs, the manager solves the following maximization problem:

∞
 
max e−rt p0 e−gt n(t)α − w · n(t) dt.
n
0

Assuming that n is continuous, the solution is to set n(t) = nM (t) at every date, where
  1
αp0 e−gt 1−α
nM (t) = . (2)
w

5 That the wage is constant amounts to a normalization, picking labor as the numeraire good in each country.
6 Samaniego (2006a) argues that, empirically, firing costs are lifted upon exit. This assumption greatly simplifies the
presentation of the paper: however, the working paper version (available upon request) shows that all results continue to
hold when firing costs must also be paid upon exit.
Section 3 introduces a flow of entrants: hence, a positive exit rate ζ is necessary for the environment to be stationary.
228 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

Here M stands for “myopic”: there are no dynamic considerations to the decision. The func-
tion nM will be useful later. Observe that, the higher the rate of technical change g, the faster the
plant falls behind the technological frontier for its industry, and the more rapid the decline in the
optimal scale of production.
More generally, let V (p0 , w; κ, g) equal the value of opening an establishment as a function of
prices p0 and w; of labor market institutions κ; and of technological parameter g. The manager’s
problem can be written:

∞
  
V (p0 , w; κ, g) = max e−rt p0 e−gt n(t)α − wn(t) + κw min 0, u(t) dt, (3)
n0 ,u
0
ṅ(t) = u(t),

t
n(t) = n(0) + u(t) dt,
0
n(0) = n0 ,
which has the appearance of a standard optimal control problem, aside from the fact that initial
employment n0 is free. In fact, the integrand is not continuously differentiable, so that Pontrya-
gin’s Maximum Principle cannot be applied directly: however, there exists a simple reformulation
of the problem that enables its application. All proofs may be found in Appendix A.
Observe that the myopic maximum nM is no longer the solution to this problem. Along nM ,
the marginal loss of profits from a variation that decreases n at any date is zero, whereas there
would be marginal savings of κw because, at each date, ṅ would be lower.

Proposition 1. The solution n∗ is characterized by a (unique) time T ∗ > 0 such that

(1) n∗ (t) = nC (T ∗ ) for t
T ∗ ;

(2) n∗ (t) = nC (t) for t > T ∗ ,

αp0 e −gt 1
where nC (t) = ( (1−rκ)w ) 1−α > nM (t).

Proposition 2. n∗ (0) < nM (0).

Proposition 3. T ∗ is increasing in κ, decreasing in g, and is constant in p0 and w.

Propositions 1 and 2 imply that, in the presence of firing costs, plants begin inefficiently small.
They hold their level of employment constant until a certain age T ∗ , after which they begin to
fire. However, by this point they are inefficiently large, and will remain so for the remainder of
their lifetimes even though they are shrinking. Interestingly, firing begins sooner in industries in
which g is high.
Several factors determine the selection of T ∗ , the age at which firing begins. Observe that T ∗
is negatively related to the constant level of employment the plant maintains initially. First, if T ∗
is too low, the plant spends too long with a level of employment above that which is myopically
efficient. Second, if T ∗ is too high, it spends too long with a level of employment below the
myopic maximum. Third, waiting longer defers the impact of firing costs into the future, as well
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 229

Fig. 2. Employment profile under firing costs.

as decreasing the amount of firing that must take place. T ∗ is the unique age that exactly balances
these factors.
Figure 2 illustrates a numerical example. The general shape of the hiring profile reflects Pro-
position 1, whereas, the result of Proposition 2 is visible in that initial employment is lower when
there are firing costs.
The key result of the paper is Proposition 4, which characterizes the relationship between
parameters and the value of entry. It implies that, ceteris paribus, EPL is most detrimental to the
profits of firms in industries in which technical change is rapid.
∂V (p,w;κ,g)
Proposition 4. The value of a plant is decreasing in the firing cost: ∂κ < 0. Moreover,
2
this decrease is the steepest when g is high: ∂ V (p,w;κ,g)
∂κ∂g < 0.

3. EPL and industry composition

In a model in which entrepreneurs have a choice over which industry to enter, Proposition 4
suggests that the presence of dismissal costs will particularly discourage entry into industries in
which technical change is rapid. We now turn to the implications of this result for cross-country
industry structure. To do so, this section extends the basic model to allow for entry and for
international trade.

3.1. Entry and trade

There are two industries. Each industry i ∈ {1, 2} is characterized by a parameter gi , the
industry rate of technical change, where g1 < g2 . There is also a set of countries C: countries
230 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

differ with regard to their institutions. In particular, they differ in terms of EPL intensity, so that
the firing cost in country c ∈ C is κc . The wage wc may also vary across countries. Product
markets are competitive, and there are no barriers to international trade. Hence, the price piτ of
good i does not vary across countries. Producers in each country take world prices as given.
In each country there is a population of entrepreneurs of measure 1. At each date, an entrepre-
neur controls a quantity 1 of an entrepreneurial resource. If eic  0 units of this input are used
to start plants in any particular industry i in country c, the output of new establishments is given
by a function k(eic ), where k  > 0, k  < 0 and limeic →+ 0 k  (0) = ∞. Entrepreneurs are active in
whichever industry is the most profitable; this is consistent with the evidence of Lazear (2004),
who finds that entrepreneurs tend not to be specialists, but rather to have broad skills that are
applicable to different fields and activities. Thus, each period, rates of entry into each industry
depend on the choices of entrepreneurs, whereas exit is governed by parameter ζ . Agents may
not create establishments in other countries: this captures the “home-bias” in equity investment
that has been well documented in the international finance literature—see French and Poterba
(1991) inter alia. Although not strictly necessary, this assumption simplifies the exposition, and
is also useful for showing that, perhaps surprisingly, differences in industry composition need
not be driven by international capital mobility.7

3.2. Industry composition

Again, assume that piτ = pi0 e−gi τ . In the notation of previous sections, the value of opening
an establishment in industry i in country c is V (pi0 , wc ; κc , gi ).
At each date, entrepreneurs solve the following problem:

Π(κc , wc ) = max V (pi0 , wc ; κc , gi )k(eic ), (4)
{eic }i
0
i

s.t. eic
1.
i

Definition 1. An entry equilibrium is a set of levels of entry {eic } and values V such that

(i) establishments are operated optimally, i.e. problem (3) is solved;

(ii) the entrepreneurial resource is used optimally in each country, i.e. problem (4) is solved.

Proposition 5. There exists a unique entry equilibrium. Moreover, for any two countries c and d,
κc < κd implies e2c > e2d .

Proposition 5 implies that the industry with the higher rate of technical change will command
a smaller share of the economy in countries where firing costs are high. Identifying this industry
with ICT, the proposition provides an account of the differences in ICT prevalence observed

7 The assumptions of strictly decreasing returns to entrepreneurship and the Inada condition are made in order for
there to be entry into all industries in all countries. The related model of Mitchell (2002) has a fixed entry cost E, and
market clearing equates profits across industries. However, in the present context, international trade implies that prices
will be determined by global rather than local conditions. When there is cross-country variation in firing costs, profits
will inevitably vary across industries and countries. Hence, a model with a fixed entry cost will have little to say about
cross-country industry structure.
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 231

in Fig. 1: since firing costs decrease profits, and since this effect is most pronounced in industries
in which technical change is rapid, countries with high levels of employment protection will tend
to specialize in other industries.

4. Learning and technical change

Are the above results are robust to extensions with more complex life-cycle dynamics? The
most salient feature of the model above is that the optimal scale of production declines over the
life cycle. Bahk and Gort (1993) estimate a plant-level production function, and they do indeed
find evidence of such a vintage effect. However, Dunne et al. (1989) find that employment tends
to grow when plants are young—a phenomenon widely attributed to establishment-level learning.
This section shows that the presence of learning effects does not affect the results.
Learning enters the production function as a new age-contingent factor L(·), so that output yτ
at date τ is

yτ = L(τ − v)evg nατ ,

where, once more, v is the vintage of the plant’s technology. Assume as before that pτ = p0 e−gτ ,
and let t = τ − v be the age of the plant. Thus, revenue at age t will be p0 L(t)e−gt nαt .

Assumption 2. The lifetime marginal revenue product profile L(t)e−gt is differentiable and
“single-peaked”—i.e. there exists some T = arg maxt L(t)e−gt such that
−gt
(i) for all t < T , ∂L(t)e
∂t > 0; and
−gt
(ii) for all t > T , ∂L(t)e
∂t < 0.

Assumption 2 implies that the productivity profile over a plant’s lifetime rises to a peak and
declines thereafter. Establishment dynamics during a plant’s youth primarily reflect the effects of
learning but, as it ages, learning wears off and the vintage effect eventually dominates.
−gt 1
Redefine the myopic maximum nM (t) ≡ ( αp0 L(t)e w ) 1−α . Also, define
  1
αp0 L(t)e−gt 1−α
n (t) ≡
C
> nM (t).
w(1 − rκ)
The following result extends Proposition 1 to the case with learning:

Proposition 6. The solution n∗ is characterized by T 0  0 and T ∗ > T 0 such that either:

(1) T 0 = 0;
(2) n∗ (t) = nC (T ∗ ) for T 0
t
T ∗ ; and
(3) n∗ (t) = nC (t) for t > T ∗ ,
or
(1) n∗ (t) = nM (t) for t
T 0 ;
(2) n∗ (t) = nM (T 0 ) = nC (T ∗ ) for T 0 < t
T ∗ ; and
(3) n∗ (t) = nC (t) for t > T ∗ .
232 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

Proposition 7. T ∗ is decreasing in g and is constant in p0 and w.

There are two possibilities, depending on parameters. One is that n∗ increases along the
undistorted path until a point at which it becomes flat. Eventually it begins firing but, as in
Proposition 1, it remains inefficiently large for the remainder of its lifetime. The other possibility
is that, if the learning curve is not steep, the solution is degenerate in the sense that T 0 = 0: in
this case, there is no strictly increasing range, and the solution is of the same form as that of the
simpler model without learning.
Interestingly, it can be shown that:

Proposition 8. T 0 < T .

In other words, immediately after the point where the plant stops hiring, it is inefficiently
small. This is a generalization of Proposition 2 to the case with learning.
An example is illustrated in Fig. 3. The overall shape of the employment profile reflects Pro-
position 6. On the other hand, the fact that the hiring profile flattens out below the productivity
peak is a consequence of Proposition 8.
∂V (p,w;κ,g)
Proposition 9. The value of a plant is decreasing in the firing cost: ∂κ < 0. Moreover,
2
this decrease is steepest when g is high: ∂ V (p,w;κ,g)
∂κ∂g < 0.

Proposition 9 implies that, in equilibrium, firing costs suppress industries in which the rate of
technical change is high under the same conditions as before, so that the industry composition

Fig. 3. Employment profile under firing costs in the presence of learning.

 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 233

result of Proposition 5 follows directly. What matters is not that there is a declining trend in
marginal revenue product (MRP) over the entire life cycle, but rather that, in industries in which
g is high, the derivative of the MRP with respect to time is lower.

5. Quantitative implications: an illustration

The previous sections show that EPL could theoretically contribute to the existence of interna-
tional differences in the diffusion of ICT. The question remains as to whether or not EPL might be
able to account for a quantitatively significant portion of the observed differences. This section
addresses the circumstances under which this might be the case with an illustrative numerical
example.
First, suppose that there are I > 2 industries. The parameters to be selected are α, ρ, ζ , k,
wc , {gi }i∈I , {pi0 }i∈I and κc . I choose k(e) = k̄eβ , and set8 α = 0.64 and ρ = 0.07, which are
standard values in the macroeconomic literature. Dunne et al. (1989) find that the 5-year hazard
rate of plants in the US is 36.3%, which yields ζ = 0.086. k̄ and wc can be any arbitrary positive
numbers, as they do not affect the optimal entry strategy (see the proof of Proposition 5): I set
them to equal one.
I interpret gi as corresponding most closely to the notion of investment-specific technical
change directed towards the capital goods used in a given industry. Cummins and Violante (2002)
report industry-level values of this rate for the United States between 1948 and 2000. I work with
10 industries, with gi set to equal between zero and 10% per year, which is the range they find.
I choose a price vector {pi0 }i∈I to ensure that nominal shares are equal across industry groups
in the benchmark economy. Finally, I examine the sensitivity of results to different values of
β ∈ (0, 1). The value of β turns out to be critical for the quantitative implications of the model.
How to calibrate labor market institutions? The empirical literature on job security often im-
plements dismissal cost using mandated severance pay applied to workers with 10 years of tenure.
Lazear (1990) and Addison and Teixeira (2003) find that, across the OECD, mandated severance
pay varies from zero to 16 months’ wages, so I consider this to be the approximate range of
interest. There are very few job security provisions in the US, so I use κc = 0 as a benchmark
as in Hopenhayn and Rogerson (1993). I compare the behavior of the benchmark economy to an
economy in which κ = 1, i.e. there is a firing cost of one year’s wages. This is towards the upper
end of the empirically relevant range, applicable to countries such as Italy, Portugal and Spain,
and is the value commonly used in quantitative exercises, such as in Hopenhayn and Rogerson
(1993). Notably, these are also the countries in which ICT is least prevalent—see Fig. 1.
The relationship between κc and the aggregate composition of the economy is reported in
Table 1, for a variety of values of β. The industry composition result of Proposition 5 is clearly
reflected in that there is a threshold industry such that, for all industries in which the rate of
technical change gi is higher, the share of output is smaller than in the benchmark economy. In
particular, this implies that the industry in which gi is the highest will have a share of GDP that
is negatively related to firing costs.
Notably, a “digital divide” of arbitrary width can be generated with an appropriate choice
of β. As β → 1, model outcomes approach those of the typical entry setup with a fixed startup

8 The working paper version of the paper shows that, for the case in which k(x) = k̄x β , Proposition 5 holds for any
number of industries, in that the relative share of any industry compared to another with lower gi decreases with κc .
This functional form is parametrically sparse, and parameter β lends itself well to interpretation as the curvature of k,
the intensity of decreasing returns to entrepreneurship, or the cross-industry sensitivity of entrepreneurship.
234 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

Table 1
Nominal shares of GDP for the benchmark economy relative to the economy with firing costs (κc = 1), for different
values of β
β Industry rate of technical change, gi (%)
1 2 3 4 5 6 7 8 9 10
0.25 −1 −1 −1 −0 −0 0 0 1 1 1
0.5 −4 −3 −2 −1 −0 0 1 2 3 4
0.75 −11 −8 −5 −3 −1 2 4 6 9 11
0.90 −27 −21 −14 −8 −1 7 14 33 31 40
0.93 −37 −28 −19 −10 1 12 24 37 51 68
0.95 −47 −36 −24 −11 4 20 39 61 86 116
0.97 −61 −47 −30 −8 19 53 96 151 221 312
Numbers are reported as percentage differences.

cost, in that all entry takes place in the most profitable industry. Hence, the higher the value of β,
the more sensitive eic will be to small cross-industry variations in profits. Thus, one can think of
β ∈ (0, 1] as a measure of intersectoral capital mobility.
This interpretation suggests that the empirically relevant value of β is best thought of in terms
of the time frame. Over a short period of time one can think of entry as not being very responsive
to variations in profits; it is unlikely that the industry composition of the US varies significantly
on a weekly, quarterly or even annual basis. At the same time, a period of 10–20 years is clearly
long enough for industry composition to change significantly, as seen in the recent surge of
ICT in the United States. Hence, this is the time-frame over which cross-industry incentives
towards entry are likely to affect the composition of GDP. Holzmann et al. (2003) document that
most EPL regimes in the OECD have been in place for several decades—long enough to have
potentially influenced industry composition. This suggests that the contribution of EPL towards
the differences in ICT diffusion could be significant: the critical factor is the long run intersectoral
mobility of entrepreneurial resources.

6. Concluding remarks

The paper shows that industry composition constitutes a channel through which EPL, and
perhaps other policies, might affect aggregates. A fairly standard industry model of technical
change is found to be consistent with the existence of a negative relationship between EPL and
ICT. As plants fall away from the technological frontier for their industry they optimally begin
to contract, and the rate at which this occurs depends on the rate of technical change. Hence,
EPL should be most costly in industries in which technical change is rapid—such as ICT. The
model predicts that the industry composition of a country with strong job security policies will
be skewed away from ICT, and skewed towards industries in which technical change is slower.
This prediction is consistent with the empirical relationship between ICT and EPL.
The model may be usefully extended in several ways. For example, plants could be allowed to
respond to the fact that their technology is determined by vintage by updating their technology
periodically. To the extent that updating is a factor of plant dynamics, firms may be able to reduce
firing by updating more frequently. This would weaken the results of the paper, and it would be
interesting to extend the model by introducing this feature. At the same time, the empirical pattern
of plant-level investment suggests that a large proportion of plants do not significantly change
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 235

their technology over their lifetimes, and that any updating occurs at widely-spaced intervals.9
Another interesting extension would be to allow firms to operate multiple plants: in this case,
they might avoid firing costs by shifting workers between plants as they fail. This extension
might require addressing the question of why some firms integrate whereas others do not.10
Finally, the analysis of the paper takes place in a partial equilibrium environment. The effect of
firing costs on the time path of prices and wages could be addressed by incorporating the model
into a general equilibrium framework.
The results beg several questions. First, might industry composition be a new and important
channel through which EPL could affect macroeconomic and labor market performance? Nu-
merical experiments suggest that this may be the case, although it naturally hinges upon the ease
with which entrepreneurial resources can shift between sectors. Second, might variations in the
rate of embodied technical change at the aggregate level interact with EPL to produce changes in
employment over time? Greenwood et al. (1996) find that, since the mid-1970s, the rate of em-
bodied technical change appears to have accelerated significantly. If this is the case, the model
suggests that differences in employment between countries with different levels of EPL might be
exacerbated, providing a novel account of the European Unemployment dilemma documented
by Ljungqvist and Sargent (1998), inter alia. Samaniego (2005) addresses this question in a gen-
eral equilibrium context. Third, if the rate of technical change is a microeconomic determinant
of industry location and comparative advantage, then countries might be destined to different
medium- or long-run growth rates based on the effect that their labor market and other institu-
tions have upon sectoral composition. The literature on barriers to growth related to for example
Parente and Prescott (2000) tends to focus on the cost of importing capital or monopoly as factors
behind low growth. This paper suggests that labor market regulation could be another factor.

Acknowledgments

I am extremely grateful to the Associate Editor and to an anonymous referee for their insights,
which substantially streamlined the presentation and proofs. All remaining errors are the author’s.

Appendix A. Proofs

First, some preliminaries. Consider the following reformulation of the problem:

∞
 
V (p0 , w; κ, g) = max e−rt p0 e−gt n(t)α − wn(t) − κwf (t) dt, (5)
n0 ,h,f
0
ṅ(t) = h(t) − f (t) dt,
h(t), f (t)  0,
n(0) = n0 . (6)

9 Doms and Dunne (1998) find that most plant level investment occurs in widely-spaced “lumps” that occur on average
once every 6 years or more. Dunne et al. (1989) find that about 40% of plants do not appear to survive for 5 years
and, since their data is quinquennial, this is likely to underestimate the true hazard rate. Identifying these “lumps” with
significant changes to the production technology (see Samaniego, 2006b) suggests that most plants may not reach that
stage.
10 See Kim (1998) for a study of the determinants of multi-unit activity in US manufacturing.
236 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

Problem (5) has two control variables, h and f . Constraints (6) imply that f (t) = 0 when the
plant is hiring, and h(t) = 0 when it is firing.
The employment path that solves problems (3) and (5) will be the same, and the objective
function in (5) is continuously differentiable in all of its arguments so that Pontryagin’s Max-
imum Principle may be applied. This reformulation is useful for any problem that involves
adjustment costs in continuous time.
Feasible employment paths are drawn from the set of piecewise-continuous and differentiable
functions. The Hamiltonian is
 
H (t) = p0 e−gt n(t)α − wn(t) − κwf (t) + µ(t) h(t) − f (t) . (7)
The first-order conditions are then:
µ(t)
0, (= 0 if h(t) > 0),
−κw
µ(t), (= µ(t) if f (t) > 0), (8)
dµ
p0 e−gt αn(t)α−1 − w = rµ(t) − , (9)
dt
µ(0) = 0, (10)
−rt
lim e µ(t)n(t) = 0, (11)
t→∞
where (10) follows from the fact that n(0) is unconstrained and (11) is the transversality condi-
tion.
For the case with learning, the Hamiltonian is
 
H (t) = p0 L(t)e−gt n(t)α − wn(t) − κwf (t) + µ(t) h(t) − f (t) . (12)
The first-order conditions are then:
µ(t)
0, (= 0 if h(t) > 0),
−κw
µ(t), (= µ(t) if f (t) > 0), (13)
dµ
p0 L(t)e−gt αn(t)α−1 − w = rµ(t) − , (14)
dt
µ(0) = 0, (15)
lim e−rt µ(t)n(t) = 0. (16)
t→∞

Lemma 1. There is no range over which the solution is strictly increasing.

Proof of Lemma 1. Suppose h(t) > 0 for t ∈ [t1 , t2 ]. Then, over this range, µ(t) = 0. From
(9) it follows that w = p0 e−gt αn(t)α−1 , for t ∈ [t1 , t2 ]. Then, n is decreasing over this interval,
which is a contradiction. 2

Lemma 2. The solution has no jumps and is not strictly monotonic, nor constant.

Proof of Lemma 2. That the solution has no jumps stems from strict concavity of the objective
function: see Arrow and Kurz (1970).
As for monotonicity, if f (t) > 0 for all t then −κw = µ(t) for every t. Since µ(0) = 0 and µ
is continuous, this is a contradiction. As for not being constant, suppose n(t) = n̄ for t ∈ [t1 , ∞).
Then, for large t, dµ/dt  rµ(t) + w  w(1 − rκ) > 0. Then, µ goes to infinity—which is a
contradiction, since µ(t)
0. 2
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 237

Lemma 3. Suppose that f (t) > 0 for t ∈ [t1 , t2 ). Then, f (t) > 0 for all t > t1 .

Proof of Lemma 3. Suppose not. Assume without loss of generality that f (t) = 0 for t ∈ [t2 , t3 ],
and that f (t) > 0 for t ∈ (t3 , t4 ]. Then for t ∈ (t2 , t3 ],
dµ
p0 e−gt2 αn(t2 )α−1 − w = −rκw > p0 e−gt αn(t)α−1 − w = rµ(t) − .
dt
Hence, dµ/dt > r[µ(t) + κw]  0, for every t ∈ (t2 , t3 ), so that µ(t3 )  −κw. But µ(t) = −κw
for t ∈ (t3 , t4 ]—a contradiction, since µ cannot jump. 2

Proof of Proposition 1. This is a corollary of the above lemmata. Also, observe that for t  T ∗ ,
we have that p0 e−gt αn(t)α−1 = w(1 − rκ) ⇒ n(t) > nM (t). 2

Proof of Proposition 2. Since µ(0) = 0, it follows that p0 αn(0)α−1 − w = −dµ(0)/dt > 0.

Hence, n(0) < nM (0). 2

Proof of Proposition 3. The conditions that T ∗ , n(0), and µ(t) (for 0
t
T ∗ ) must solve are
the following:
∗
p0 e−gT αn(0)α−1 = w(1 − rκ), (17)
dµ
= rµ(t) + w − p0 e−gt αn(0)α−1 , (18)
dt
µ(0) = 0, (19)
∗
µ(T ) = −κw. (20)
Solving the differential equation (18) with terminal conditions (19) and (20), and using (17)
yields
r(κw)2 w(1 − rκ)
gT ∗ 
−κw − = wT ∗ − e −1 ,
2 g
which yields T ∗ . The left-hand side is constant in g, whereas the right-hand side is decreasing
in g. Thus, T ∗ decreases. On the other hand, when κ increases, the right-hand side shifts up and
the left-hand side shifts down. Thus, T ∗ increases. From Eq. (17) we see that the effects on n(0)
are opposite to those on T ∗ . In fact,
∂T ∗ T∗
= −(r+g)T ∗ < 0. 2
∂g ge −g
Proof of Proposition 4. Applying the Envelope theorem to the optimized control problem yields

∞  αp 1 ∗
∂V 0 1−α − gT
= −w f (t) dt = − e 1−α .
∂κ 1 − rκ
0
Finally, it can be shown that
 −(r+g)T ∗ 
∂ 2V T ∗  αp0 1−α
1
gT ∗ e
= e− 1−α −(r+g)T ∗ < 0.
∂κ∂g 1 − α w − rκw e −1
That n∗ (0) < nM (0) stems from earlier results. 2
238 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

Proof of Proposition 5. Optimal use of the entrepreneurial resource implies that, in any coun-
try c, V (pi0 , wc ; κc , gi )k  (eic ) = V (pj 0 , wc ; κc , gj )k  (ej c )∀i, j . Before anything else, it is use-
1/(1−α) −α/(1−α)
ful to note that V (p0i , wc ; κc , gi ) = p0i wc V (1, 1; κc , gi ). Hence, entry will be set
1/(1−α) −α/(1−α) 
so that p0i wc V (1, 1; κc , gi )k (eic ) = Mc ∀i, where Mc is the (endogenous) value of
entry in country c relative to the wage, and which will be determined by market clearing. As a
V (1, 1; κc , gi )k  (eic ) will be equal across industries. Let Vic = V (1, 1; κc , gi ).
1/(1−α)
result, p0i
Consider d : κd > κc . We know that in country c, p0i
1/(1−α) c , where M
Vic k  (eic ) = M c =
−α/(1−α) 1/(1−α)  d . Moreover, let the interval
Mc /wc . Proposition 4 implies that p0i Vid k (eic ) < M
 1 1
U = p01 1−α
V1d k  (e1c ), p02 1−α
V2d k  (e2c ) .

It has to be that M d ∈ U . Now consider any u ∈ U , and pick eid such that p 1/(1−α) Vid k  (eid ) = u.
 0i
It may not bethe case that i eid = 1—indeed there is one and only one such u for which this is
the case, as i eid will be decreasing in u. Nonetheless, for any u, strict concavity of k implies
that there is only one eid that satisfies this equation.
Let u (unregulated) and r (regulated) be countries such that κr > κu . In equilibrium, en-
V2c k  (e2c ) =
1/(1−α)
try in either industry should yield the same marginal return, so that p2
V1cu k  (e1c ). Rearranging, we have that
1/(1−α)
p1
1
p21−α V2c k  (e1c )
= .
1
k  (e2c )
p11−α V1c
V2j (1)
First note that V2j < Vlj for any country j , so that Vlj (1) < 1. Second, Proposition 4 implies
V2r V2u
that V1r < V1u . This, along with
1
p21−α V2c k  (e1c )
= ,
1
1−α k  (e2c )
p1 V1c
 k (e1u )  k  (1−ehr ) k  (1−ehu )
implies that kk  (e 1r )
(e2r ) < k  (e2u ) . Finally, eaj = 1 − ecj in any country j , so that k  (ehr ) < k  (ehu ) .
What does this say about e2r vs. e2hu ? Applying the product rule,

d k k(1−e)
−k  (1 − e)k  (e) − k  (e)k  (1 − e)
 (e)
= > 0.
de k  (e)2
Consequently, e2u > e2r and e1u < e1r . 2

Lemma 4. If n∗ is strictly increasing, n = nM . This cannot occur for t > T .

Proof of Lemma 4. Suppose h(t) > 0 for t ∈ [t1 , t2 ]. Then, over this range, µ(t) = 0. From (14)
it follows that w = p0 L(t)e−gt αn(t)α−1 , for t ∈ [t1 , t2 ]. Rearranging yields the result. 2

Lemma 5. If n∗ is strictly decreasing, n = nC . This cannot occur for t < T .

Proof of Lemma 5. Suppose f (t) > 0 for t ∈ [t1 , t2 ]. Then, over this range, µ(t) = −κw. From
(14) it follows that w(1 − rκ) = p0 L(t)e−gt αn(t)α−1 , for t ∈ [t1 , t2 ]. Rearranging yields the
result. 2
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 239

Lemma 6. Suppose that f (t) > 0 for t ∈ [t1 , t2 ). Then, f (t) > 0 for all t > t1 .

Proof of Lemma 6. Suppose not. Assume without loss of generality that f (t) = 0 for t ∈ [t2 , t3 ],
and that f (t) > 0 for t ∈ (t3 , t4 ]. Then for t ∈ (t2 , t3 ], p0 L(t2 )e−gt2 αn(t2 )α−1 − w = −rκw >
p0 L(t)e−gt αn(t)α−1 − w = rµ(t) − dµ/dt, where the inequality follows from the fact that
(t2 , t3 ] is in the range over which L(t)e−gt is decreasing in t, since this has to be true of
t ∈ [t1 , t2 ) (by Lemma 5). Thus, dµ/dt > r[µ(t) + κw]  0, for every t ∈ (t2 , t3 ). Consequently,
µ(t3 )  −κw. But µ(t) = −κw for t ∈ (t3 , t4 ]. This is a contradiction, since µ cannot jump.
Moreover, the equations imply that while f (t) > 0, n(t) = nC (t). This proves that it cannot
suddenly increase over a range and then decrease again, since n cannot jump. 2

Proof of Propositions 6 and 7. Corollaries of the above lemmata. 2

Proof of Proposition 8. The derivative of the objective function with respect to T reduces to the
expression

T


e−rt uC (T )π  nC (T ) dt + e−rT κuC (T ),
G(T )

where uC is the derivative of nC . Observe that uC (T ) < 0, and, for all t > T , π  (nC (T )) < 0
where π is the instantaneous profit function net of firing costs. Thus, the value of this expression
for G(T ) = T is positive: T ∗ = G−1 (T ) is not the solution. 2

Proof of Proposition 9. As before, the Envelope theorem implies that

∞
∂V g  αp0 1−α
1
  1
=− × L(t)e−gt 1−α dt,
∂κ 1 − α 1 − rκ
T∗

which is negative. From there,

∂ 2V  αp 1 T ∗ g  ∗  1 ∂T
∗
× L(t)e−gT 1−α
0 1−α
=− ,
∂κ∂g 1 − rκ 1−α ∂g
so that a necessary and sufficient condition for ∂ 2 V /(∂κ∂g) to be negative is that ∂T ∗/∂g < 0.
The remainder of the proof demonstrates that this property does indeed hold. There are two cases.
If κ is so high that the plant does not grow upon birth, then the problem is of the same form as in
the absence of learning and the result follows from there. The more interesting case is when the
plant does grow for a period before it ceases to hire. Proposition 6 implies that T ∗ is given by

)
G(T
T
∗ −rt



T = arg max e π nM (t) dt + e−rt π nC (T ) dt
T
0 G(T )

∞


+ e−rt π nC (t) + e−rt κwuC (t) dt,
T
240 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241

where G(T ) = t : nM (t) = nC (T ) and π is the profit function, appropriately specified. G(T ) is
unique because of single-peakedness, and G(T ) < 0 for the same reason. The first order condi-
tion of this problem reduces to
   
1 − rκ −(r+g)[G(T ∗ )−T ∗ ] 1 − rκ ∗ ∗ gκ
0= ∗
− 1 − e ∗
− e−r[G(T )−T ] − .
αL(T ) αL(T ) 1−α
Taking the total derivative with respect to g,
 
∂T ∗   ∗  −(r+g)[G(T ∗ )−T ∗ ] 1 − rκ
0= (r + g) G (T ) − 1 × e
∂g αL(T ∗ )
 
 ∗ ∗
 −(r+g)[G(T ∗ )−T ∗ ] 1 − rκ
+ G(T ) − T × e
αL(T ∗ )
∗
∂T    ∗ ∗ κ
+ r G (T ) − 1 × e−r[G(T )−T ] − .
∂g 1−α
Observe that G(T ) < T so that all the expressions in this equation that are not multiplied by
∂T ∗/∂g are negative—as are the expressions that are multiplied by ∂T ∗/∂g. Consequently, for
this equation to hold, ∂T ∗/∂g < 0. 2

References

Addison, J.T., Teixeira, P., 2003. What have we learned about the employment effects of severance pay? Further iterations
of Lazear et al. Mimeo. University of South Carolina.
Arrow, K., Kurz, M., 1970. Public Investment, the Rate of Return, and Optimal Fiscal Policy. Johns Hopkins Press,
Baltimore.
Autor, D.H., Levy, F., Murnane, R.J., 2003. The skill content of recent technological change: An empirical exploration.
Quarterly Journal of Economics 118, 1279–1333.
Bahk, B.-H., Gort, M., 1993. Decomposing learning by doing in new plants. Journal of Political Economy 101, 561–583.
Bartelsman, E.J., Gray, E., 1996. The NBER manufacturing productivity database. Technical working paper 205. NBER,
Cambridge.
Colecchia, A., Schreyer, P., 2002. ICT and economic growth in the 1990s: Is the United States a unique case? Review of
Economic Dynamics 5, 408–442.
Coppel, J., 2000. E-commerce: Impacts and policy changes. Working paper 2000-25. OECD Economics Department.
Cummins, J.G., Violante, G.L., 2002. Equipment-embodied technical change in the US (1947–2000): Measurement and
macroeconomic consequences. Review of Economic Dynamics 5, 243–284.
Doms, M., Dunne, T., 1998. Capital adjustment patterns in manufacturing plants. Review of Economic Dynamics 1,
409–429.
Dunne, T., Roberts, M.J., Samuelson, L., 1989. The growth and failure of US manufacturing plants. Quarterly Journal of
Economics 104, 671–698.
French, K.R., Poterba, J.M., 1991. Investor diversification and international equity markets. American Economic Review
(Papers and Proceedings of the Hundred and Third Annual Meeting of the American Economic Association) 81,
222–226.
Gordon, R.J., 1990. The Measurement of Durable Goods Prices. Univ. of Chicago Press, Chicago.
Greenwood, J., Hercowitz, Z., Krusell, P., 1996. Long-run implications of investment-specific technological change.
American Economic Review 87, 342–362.
Gust, C., Marquez, J., 2004. International comparisons of productivity growth: The role of information technology and
regulatory practices. Labour Economics 11, 33–58.
Holzmann, R., Kripa, I., Vodopivec, M., 2003. Severance pay programmes around the world: Rationale, status and re-
forms. Mimeo. World Bank.
Hopenhayn, H., Rogerson, R., 1993. Job turnover and policy evaluation: A general equilibrium analysis. Journal of
Political Economy 101, 915–938.
Kim, S., 1998. The rise of multi-unit firms in US manufacturing. Working paper 6425. NBER.
Lazear, E.P., 1990. Job security provisions and employment. Quarterly Journal of Economics 105, 699–726.
 R.M. Samaniego / Review of Economic Dynamics 9 (2006) 224–241 241

Lazear, E.P., 2004. Balanced skills and entrepreneurship. American Economic Review 94, 208–211.
Ljungqvist, L., Sargent, T.J., 1998. The European unemployment dilemma. Journal of Political Economy 106, 514–550.
Mitchell, M.F., 2002. Technological change and the scale of production. Review of Economic Dynamics 5, 477–488.
Nicoletti, G., Scarpetta, S., Boylaud, O., 2000. Summary indicators of product market regulation with an extension to
employment protection legislation. Working paper 226. OECD Economics Department.
Oliner, S.D., Sichel, D.E., 2000. The resurgence of growth in the late 1990s: Is information technology the story? Journal
of Economic Perspectives 14, 3–22.
Parente, S.L., Prescott, E.C., 2000. Barriers to Riches. MIT Press, Cambridge.
Pilat, D., Lee, F., 2001. Productivity growth in ICT-producing and ICT-using industries. Working paper 2001-4. OECD
Economics Department.
Samaniego, R., 2005. Can technical change exacerbate the effects of labor market sclerosis? Mimeo. George Washington
University.
Samaniego, R., 2006a. Do firing costs affect the incidence of firm bankruptcy? Macroeconomic Dynamics. In press.
Samaniego, R., 2006b. Organizational capital, technology adoption and the productivity slowdown. Journal of Monetary
Economics. In press.
Wilson, D.J., 2002. Is embodied technology the result of upstream R & D? Industry-level evidence. Review of Economic
Dynamics 5, 285–317.