exports, innovation and production growth: a dynamic ... · exports, innovation and production...
TRANSCRIPT
Exports, Innovation and Production Growth:
A Dynamic Heterogeneous Firm Model with Learning and
Entry Costs
Helen Ruohan Wu∗†
The Ohio State University
November 13, 2012
Abstract
In this paper, I analyze a dynamic model of a firm’s joint decisions to export and
innovate, allowing both decisions to affect the firm’s production in accordance with self-
selection and learning-by-exporting theories of firm-level export and production dynamics.
I calibrate the model using Chilean manufacturing plant data from 2005 to 2007 and
find that self-selection and learning-by-exporting alone cannot adequately explain the
observed cross-sectional relationship between firm level exports and production, favoring
instead a model that requires both mechanisms to work in tandem. Counterfactual policy
analysis indicates that the impact of trade liberalization depends on the maturity of
the industry, the rate at which liberalization occurs, and the nature of the dominant
adjustment mechanism governing firm-level export-production dynamics.
JEL Classification: C61; C82; F17; F43; F63; O14; O32.
Keywords: Exports; Production; Innovation; Self-Selection; Learning-by-Exporting;
Heterogeneity; Trade Liberalization.
∗Email: [email protected]; Address: Department of Economics, The Ohio State University, Arps Hall 410,1945 North High Streeet, Columbus, OH 43210-1172.†I am grateful to Professor Mario Miranda, Professor Paul Evans, and participants in The Ohio State
University Department of Economics Macroeconomic Theory Seminar for helpful comments on earlier draftsof this paper. I would also like to thank the National Statistics Institute of Chile for providing the data usedin this paper and Alexander Gotthard-Real for helping me to understand the data.
1 Introduction
It is well documented that firms that export are more productive than firms that do not export.
Two theories have been proposed to explain this observed phenomenon. Self-selection theory
maintains that exporting is a consequence of high firm production. Specifically, exporting
is profitable. However, to enter the export market, a firm incurs unrecoverable entry costs
and typically faces higher fixed-costs of production over time. Entry into the export market
is therefore sensible only if the firm is sufficiently productive to generate the high volume of
exports needed to cover the entry costs and the higher continuing fixed costs.
Learning-by-exporting theory, on the other hand, maintains that high firm production is
a consequence of exporting behavior. Specifically, upon entering the export market, firm
production will grow at a faster rate. Various explanations have been used to justify this
claim, including that exporters become more productive as a result of intense yet informative
global competition and that they have greater access to state-of-the-art technology.
However, exporting may not be strictly a cause nor strictly a consequence of a firm’s
production. That is, exporting and production may mutually reinforcing, in which case nei-
ther self-selection theory nor learning-by-exporting theory alone will adequately explain the
complex relationship between exports and production.
To study the relationship between exports and firm production, I develop a dynamic
optimization model of an individual firm that allows for both self-selection and learning-by-
exporting. Following Melitz (2003) [20] and Constantini & Melitz (2008) [14], the firm faces
unrecoverable export market entry costs, but enjoys faster production gains when it becomes
an exporter. Each period, the firm faces three key decisions. First, facing an unrecoverable
cost of entering the export market, the firm must decide whether to enter. Second, if the firm
enters or has previously entered the export market, it must then decide how much to export
and how much to sell in domestic markets. Third, the firm, whether an exporter or non-
exporter, must decide how much to invest in production-enhancing innovation, the benefits
of which may depend on the export status of the firm.
My model makes a novel contribution to the self-selection and learning-by-exporting lit-
eratures by allowing export and innovation decisions to be continuous. Specifically, unlike
existing models in which innovation and exportation are modeled as discrete one-time choic-
es, I allow a firm that has entered the export market to sell its product in both the domestic
and export markets to degrees that may vary over time and firm production. Innovation is
also viewed as a continuous action whose costs may depend explicitly on whether the firm is
or is not an “exporter”.
I extend my representative firm model to consider an industry consisting of many firms
that are heterogeneous with respect to age, production, and exporter status. Specifically, each
1
period, a fixed percentage of existing firms are forced to exit the industry for purely exogenous
reasons. Simultaneously, a fixed number of new, as yet non-exporting firms enter the industry
with an initial production drawn from a fixed probability distribution. I calibrate the model
using Chilean manufacturing plant data from 2005-2007 obtained from the Annual National
Industrial Survey of the National Statistics Institute of Chile (INE, Instituto Nacional de
Estadisticas Chile).
In my heterogeneous firm industry model, aggregate production, exports, and innovation
vary over time, but eventually converge to a steady state. I perform simulations to compare the
dynamic adjustments of the industry and the eventual steady-states achieved by the industry
under three distinct scenarios: a benchmark scenario that allows for both self-selection and
learning-by-exporting, a scenario that allows for self-selection but not learning-by-exporting,
and a scenario that allows for learning-by-exporting but not self-selection. I also perform
simulations to assess the impacts of trade liberalization on the industry at different stages of
industry development, ranging from an infant industry to a mature industry.
I find that self-selection and learning-by-exporting influence firm growth in different ways.
Specifically, without learning-by-exporting, self-selection leads to unrealistically low levels of
exports; without self-selection, however, all non-exporters instantly become exporters. Both
mechanisms jointly are needed to explain the stylized facts of Chilean manufacturing. I also
find that trade liberalization that raises the effective price received by exporters will have a
more profound impact on the level of exports in the short- and medium-run if the industry
relatively young than when it is mature.
This paper is organized as follows. Section 2 develops a model of a representative firm’s
joint decisions to export and innovate. Section 3 describes and summarizes the benchmark
parameter values employed in the study and the data from which the parameter estimates were
drawn. In section 4, I analyze the optimal export and innovation policies of a representative
firm under the benchmark parameterization. In section 5, I examine the two polar cases in
which either self-selection and learning-by-exporting mechanisms, but not both, are operant.
In section 6, I report findings from a counterfactual analysis of trade liberalization. Section 7
concludes the paper.
1.1 Literature Review
There is an extensive body of empirical research that supports the hypothesis that investment
in innovation and technology can reasonably explain the observed correlation between firm
production and exports. Hallward-Driemeier et al. (2002) [15], Baldwin and Gu (2004) [6],
Bustos (2005) [11], and Aw et al. (2007) [4] find that firms’ exports are related to research
and development (R&D) investment or adoption of new technology. The key insights drawn
from these studies is that innovation and exportation are correlated and that both influence
2
the firm’s production growth. Numerous theoretical studies on the role of innovation in
exporting decisions have complemented the empirical studies. Atkeson & Burstein (2010) [3]
and Constantini & Melitz (2008) [14] find that trade liberalization increases the rate of return
to a firm’s R&D investment, and thus leads to production growth. In both papers, firms
are distinguished only by their production. Wu (2011, 2012) [26] [27], using industry-level
and plant-level data, finds that significant learning-by-exporting effects are present only when
industries are investing substantial amounts in innovation.
Self-selection theory (Melitz 2003 [20]; Bernard et al. 2003 [8], 2006 [9]) has enjoyed
widespread theoretical and empirical support. Empirical tests of learning-by-exporting theory
(Marin 1992 [19]; Ben-David 1993 [7]), however, have met with mixed results. Empirical tests
based on different data sets have yielded conflicting conclusions, indicating that the theory
may be case-sensitive. For example, studies using data from Indonesia (Amiti & Konings
2004 [2]), Canada (Baldwin & Gu 2004 [6]), Britain (Blalock et al. 2004 [10]) and Chile
(Alvarez & Lopez 2005 [1]) find strong evidence that exporting firms are more productive.
However, Clerides et al. (1998) [13], using plant-level data from Mexico, Columbia and
Morocco, find no evidence that production costs are affected by entering the export market.
Using the data of Canadian plants, Lileeva et al. (2010) [18] find significant learning effect
only among those induced by the tariff cuts to start exporting or to export more. To resolve
the apparent conflict, recent studies (Bustos 2011 [12]; Verhoogen 2004 [24]; Aw, Roberts &
Winston 2007 [4]) have suggested that learning-by-exporting is possible only if firms make
sufficiently high investment in innovation.
Other recent research has compared the mechanisms hypothesized by self-selection and
learning-by-exporting theories. Wagner (2007) [25] performed a review of the empirical lit-
erature, finding strong support for self-selection theory and mixed support for learning-by-
exporting theory. He concluded that “exporters are more productive than non-exporters, and
the more productive firms self-select into export markets, while exporting does not necessarily
improve production”. Imbruno (2008) [16] tested for both mechanisms using Italian manufac-
turing firms, finding that self-selection, rather than learning-by-exporting, is responsible for
high production among exporters. Other researchers have combined both mechanisms in a
single model to study export-production dynamics. Constantini & Melitz (2008) [14] and Aw
et al. (2008) [5], for example, treat firm-level exporting and innovation as discrete all-or-none
decisions.
2 Model
The representative firm begins each period either as an established exporter i = 1 or non-
exporter i = 0, with a predetermined production q ≥ 0, namely how much a firm produces.
3
Being in a competitive market, the firm takes the domestic market price P0 and the export
market price P1 as given (hereafter, P = (P0,P1) indicates the joint price vector). The firm
must then decide whether to assume exporter status j = 1 or not j = 0 for the current period,
and, if it assumes exporter status, what proportion y of its production q to export, selling the
remainder domestically.
The firm must also decide how much to invest in innovation that would enhance productive
capacity next period. Specifically, the firm’s production in the following period
q′ = (1− γ)q + x (1)
is a function of its current production q, the amount it invests in innovation x ≥ 0, and
per-period rate of production depreciation γ ∈ (0,1).
Each period the firm faces a fixed cost κ1 if it is an exporter (j = 1) and a fixed cost κ0
if it is not an exporter (j = 0), where κ1 > κ0 > 0. A non-exporter that wishes to become
an exporter must make a fixed investment K > 0 which is unrecoverable. To economize on
notation, we let ηij denote the sum of fixed production and export market entry and exit costs
associated with the firm choosing exporter status j this period, given it begins the period in
exporter status i. Thus the export-status switching cost is:
ηij =
{κ1 +K if i = 0, j = 1
κj if otherwise.
The firm also incurs a quadratic variable production and marketing cost:
D(q, y) = d0(q(1− y))2 + d1(qy)2; (2)
and incurs a variable innovation cost:
Cj(x) = cjxβ. (3)
with β > 0. I assume that exporting production and marketing is more expensive than the
domestic one (d1 > d0 > 0) and, to capture learning-by-exporting, that an exporter faces
lower innovation costs than a non-exporter (c0 > c1 > 0).
The firm maximizes the present value of current and expected future profits over an
infinite horizon, given that it faces a probability µ of going bankrupt and ceasing to exist
at the beginning of each period. The firm’s dynamic decision problem is characterized by a
Bellman equation whose value function specifies the maximum present value of current and
expected future profits Vi(q) attainable by the firm, given its export status i and production
q at the beginning of the period:
Vi(q) = maxj=0,10≤y≤jx≥0
{P0q(1− y) + P1qy − Cj(x)−D(q, y)− ηij + δ(1− µ)Vj((1− γ)q + x)}. (4)
4
Here, δ ∈(0 1) is the firm’s per-period discount rate.
The industry is assumed to consist of many firms that behave as the representative firm,
but which differ with respect to age and initial production. Specifically, a fixed proportion µ
of firms exit the industry randomly in any period but are replaced by a fixed number of new
firms with production drawn randomly from a known distribution. Specifically, I assume that
a new firm’s initial production, or production is drawn randomly from gamma distribution
Γ(q|k, θ) with shape parameter k and scale parameter θ. Idiosyncratic variation in initial
production and time of death produces an industry with firms that are heterogenous with
respect to production and export status; the total number of firms in the industry increases
until the number of firms exiting equals the number of firms entering.
3 Data and Parameterization
Table 1 reports the benchmark values of the model parameters used in model simulation-
s. In our parameter specification, we normalize the domestic price P0 to 1 and express all
other prices and costs relative to it. The remaining model parameters can be divided into
three classes. First, parameters estimated directly from from data, which include the export
price, fixed cost of exporters and non-exporters, depreciation rate, the death probability and
the initial production distribution parameters. Second, parameters whose values are drawn
from econometric estimates, which include the non-exporters’ innovation cost parameter, the
production cost function parameters, and the innovation cost rate. Column “Range” shows
the 95% credit interval of the estimation results. Third, parameters that are chosen through
calibration, that is, chosen so that simulated values of prescribed endogenous variables match
empirically observed values; these parameters include exporters’ innovation cost, and the
export market entry cost.
To parameterize the model, I use 2005-2007 Chilean manufacturing plant data obtained
from the National Industrial Survey conducted by the Chilean National Statistics Institute
(INE, Instituto Nacional de Estadisticas). The annual INE survey includes plants that began
operating during that year and excludes those that closed for any reason. In the survey,
production, exports, overseas business connections, and other manufacturing information are
collected at the plant level. Although firms may own more than one plant, most plants in the
survey are single-firm plants.1 Table 4 in Appendix B provides the descriptive statistics of
the observed plants.
I use the average income tax rate on firms as the fixed cost κ0 faced by a non-exporting
firm. I add the average tax rate on exports to generate an estimate the fixed cost κ1 faced
by exporting firms. To calibrate export market entry costs and exporter innovation cost
1According to Pavcnik (2002) [23], more than 90% of Chilean manufacturing firms have only one plant.
5
Table 1: Parameterization
Description Symbol Benchmark Value Range
Domestic Price P0 1.0
Export Price P1 2.0
Exporting Start-Up Cost K 1.12
Fixed Cost - Non-Exporter κ0 1.0
Fixed Cost - Exporter κ1 2.0
Innovation Cost Parameter - Exporter c1 0.18
Innovation Cost Parameter - Non-Exporter c0 0.9 0.98 ∼ 1.00
Production Cost Parameter - Exporting Products d1 0.85 1.02 ∼ 1.31
Production Cost Parameter - Domestic Products d0 0.2 0.20 ∼ 0.24
Innovation Cost Rate β 10 9.12 ∼ 12.19
Production Distribution Shape Parameter k 2.5 2.48 ∼ 2.66
Production Distribution Scale Parameter θ 0.2 0.20 ∼ 0.22
Death Probability µ 0.15 0.11 ∼ 0.17
Production Depreciation Rate γ 0.2
Discount Rate δ 0.9
parameters, I use two targets from the data: relative exporting supply, namely the ratio
between aggregate exports and domestic marketings QS1 /QS0 ; and the exporter percentage.
Among the observed Chilean manufacturing plants, the relative exporting supply is 0.19.
And among all the 6,614 observations, there are only 1,624 (25%) plants being exporters. I
then calibrate K and c1 to match both of these targets.
Other parameters estimates were drawn from Wu (2012) [27], who used an extended
Olley-Pakes method to estimate plant-level productivity from the INE data set. The Olley-
Pakes (Olley & Pakes 1996 [22]) method employs a semi-parametric algorithm to remove
simultaneity and endogeneity from the unobserved production shock and capital input. The
conventional estimation method treats investment and capital as quasi-fixed inputs; Wu (2011,
2012) [26] [27] extends estimation by including two additional variables: the export ratio and
the export growth rate. In this paper, I use the production to indicate how productive a firm
is. Thus hereby I use the firms’ estimated productivity to represent their production levels.
The death probability µ and the parameters of the new plant production Γ distribution were
directly estimated from the 2005-2007 ENIA data, which tracks the number of firms that
enter and exit each year (See Figure 18 of Appendix C). The comparison of productivities
between non-exporters and exporters are shown in Figure 19 of Appendix C. The density
6
histogram of the production distribution of the exporters’ productivity is given with a much
higher median, and a more right-skewed shape.
4 Firm-Level Simulations
I begin with an analysis of the representative firm’s optimal export and innovation policies
and the evolutionary path taken by an individual firm that follows these policies.
The Bellman equation (4) is solved using the method of collocation (Judd 1998 [17];
Miranda and Fackler 2002 [21]). The collocation method calls for the value functions Vi(q) to
be approximated using a linear combination of n known basis functions φj :
Vi(q) ≈∑
j=1...n
ςijφj(q) (5)
The 2n unknown coefficients ςij are then fixed by requiring the value function approximate
to satisfy the Bellman equation (4), not at all production levels q, but rather at n judiciously
chosen collocation nodes qj . The collocation method replaces the Bellman functional equations
with a set of 2n nonlinear equations with 2n unknowns that are solved using Newton’s method.
The collocation method can generate highly accurate approximate solutions to the Bellman
equation, provided the basis functions and collocation nodes are chosen judiciously and their
number n is set adequately high. For this paper, I chose linear orthogonal polynomial basis
functions and n = 150 linear nodes to compute the approximate solutions for the Bellman
equations. The solution was computed using the CompEcon 2010 Toolbox routine dpsolve
(Miranda and Fackler 2002 [21]).
Figure 1: Conditional Value Functions for Non-Exporter
7
4.1 Optimal Policy
Figure 1 illustrates the conditional value functions for a non-exporter under the benchmark
parameterization. The horizontal axis represents the production q, while the vertical axis
represents the value. For a current non-exporter, the value associated with becoming an
exporter is the red line, while the value of remaining a non-exporter is the blue line. The
two curves intersect each other at a critical production q∗ = 2.81 at which a non-exporter
becomes an exporter. On the left side of q∗, the value of becoming an exporter exceeds the
value of remaining a non-exporter.
Figure 2 illustrates the representative firm’s optimal innovation policy (left panel) and
optimal export policy (right panel). On both panels the horizontal axis represents production
q.
As seen in the right panel of Figure 2, if q < q∗, the firm remains a non-exporter; if
q > q∗, the firm becomes an exporter. For an exporter, the proportion of production exported
declines with production due to rising production and marketing costs.
Figure 2: Optimal Policy
As seen in the left panel of Figure 2, optimal innovation initially declines with production,
but begins to rise in an irregular manner as production approaches the critical level q∗ at
which the firm becomes an exporter. This issue is addressed specifically in Figure 3. Those 5
kinked points corresponds to 5 specific critical q’s. Point q∗5 = 1.56 means if a firm starts as a
non-exporter with a q on the left hand side of q∗5, it needs more than 5 periods to jump and
start to export. Point q∗4 means if a firm starts as a non-exporter with a q on the right side
of q∗5 and on the left of q∗4 (= 1.95), it needs more than 4 periods to jump. If it starts with a
q between q∗4 and q∗3 (= 2.23), it needs 4 periods to jump. A firm starts with a q between q∗3
and q∗2 (= 2.43) needs only 3 periods to jump; between q∗2 and q∗1 (=2.60) needs 2 periods to
jump. Once a firm starts with a q that is higher than q∗1 yet lower than critical q∗, the firm
8
will jump after only 1 period. Between each q∗m (m = 1, 2, . . . , 5) the optimal innovation will
decrease as q increases; the more productive a firm is, the less innovation it needs to make.
Figure 3: Optimal Policy of an Ex-Post Non-Exporter
At the critical production level q∗, innovation makes a discrete jump. However, for an
exporter, the level of innovation subsequently decreases with production, eventually falling
to zero as the production exceeds 4.73. For a highly-productive firm, it does not invest in
innovation anymore. Instead, the firm will let the production fall and cut back its production
and marketing expenditure. Eventually its production will converge to the steady-state value
3.59, with the steady-state innovation investment 0.72.
4.2 Individual Firm Dynamics
Consider now a firm that starts as a non-exporter with initial production q = 0.1 and lives
forever. As seen in Figure 4, the firm innovates from the outset and its production grows.
In period 9, the firm’s production reaches the critical level q∗, at which point it becomes an
”exporter”. From that point forward, the firm’s production continues to grow, but at a de-
clining rate, eventually approaching a steady-state of 3.59. Any firm, if it survives sufficiently
long, will become an exporter and approach this steady-state, regardless of its production at
inception.
The dynamics of exports and innovation for the representative firm are illustrated in Figure
5. As seen on the left panel, the firm does not export until the 9th period; once this happens,
the proportion of production exported jumps to its highest level. However, the export ratio
subsequently declines gradually as production increases because of the increasing production
cost, eventually converges to its steady-state.
9
Figure 4: Path of An Individual Firm’s Production
As seen on the right panel of Figure 5, a nascent non-exporting firm innovates, but ini-
tially at a declining rate due to rising costs of innovation. However, as a firm’s production
approaches the critical level q∗ at which the firm becomes an exporter, the firm’s production
begins to rise as the firm anticipates becoming an exporter and being able to take advantage
of the lower innovation cost that would bring. Innovation reaches the maximum in the period
the firm becomes an exporter. Once the firm becomes an exporter, its innovation begins to
decline, again due to rising marginal cost of innovation, eventually dropping to its steady-state
value, the level of innovation needed to cover depreciation of production in its steady-state.
Figure 5: Paths of An Individual Firm’s Exports and Innovation
In the long run, the distributions of the q, x and y across all the different firms will
converge to steady-states. Under the benchmark parameter setting, at the steady-state the
aggregate domestic supply is 1.80, while the aggregate exports is 0.34. The relative export
supply is 0.19, and 25% of the firms are exporters. Both targets are successfully replicated.
10
5 Self-Selection vs. Learning-by-Exporting
How does self-selection and learning-by-exporting affects firms’ and industries’ dynamics re-
spectively? What is the specific role each of them plays? To unravel the effects of self-
selection and learning-by-exporting, I now simulate the model under two extreme cases: one
with self-selection but no learning-by-exporting, and one with learning-by-exporting but no
self-selection. Previously under the benchmark model, both relative exporting supply and
exporter percentage were managed to be replicated. In this section, I will examine how well
each extreme case behaves. If I remove one mechanism from the consideration, will the model
still be able to replicate both targets? If not, what results can be attained?
5.1 Pure Self-Selection Model
I eliminate learning-by-exporting from my model by setting the innovation cost for the ex-
porters c1 and the innovation cost for non-exporters c0 equal to one another. Under this
assumption, the firm’s export market participation has no influence on its costs of innovation.
The only remaining cost consideration for a firm contemplating entering the export market is
the cost of entry associated with self-selection theory.
Next I recalibrate the model to see how it can replicate the stylized facts. I choose two
empirical targets: the relative export supply as the ratio between aggregate export supply and
aggregate domestic supply, and the exporter percentage on the entire industry. To replicate
the stylized facts, I adjust the parameterization; specifically, production cost for domestic
products d1 needs to be raised by 10%. As a result, the steady-state export ratio increases,
while innovation and production decreases. Without the learning benefit from exporting,
the innovation behavior is discouraged. Even with a higher export ratio, pure self-selection
mechanism leads to a lower production.
However, only the exporter percentage can be successfully matched with the data. Because
of the lack of the learning effect, the export supply is discouraged; even with the same exporter
percentage, the amount of the products supplied to the overseas decreases. As a result, the
relative export supply decreases by 21%. The pure self-selection can not thoroughly replicate
the stylized facts. Compared with the benchmark model, the export intensity will be greatly
discouraged in terms of less export supply. Without learning-by-exporting mechanism, the
positive correlation between exports and production can only be partially explained. As a
result, the exporting behavior of an industry will be much more shrunk, which is not what
the current globalization indicates.
Figure 16 in Appendix C depicts the simulated growth path of a firm beginning from q =
0.1. It still takes 9 periods for this firm to become an exporter, the same as in the benchmark
model. Notice that under this circumstance, the innovation investment reaches its peak at
11
the 8th periods, one period before the switching moment. Afterwards the innovation declines
gradually because of the increasing marginal cost. Production smoothly increases with a
decreasing rate, yet experience a bump at the 8th period because of the innovation peak.
Other than these differences, all the other findings are similar to the benchmark results.
Self-selection mechanism therefore separates domestic market from the export market. At
the steady-state, the export ratio rises by 6%; the innovation decreases by 8%, so does the
steady-state production. On the industry-level, the exporter percentage is replicated, however
the relative export supply decreases 21%. Without learning-by-exporting, the export inten-
sity of an industry declines. Therefore, the learning-by-exporting mechanism can effectively
encourage the export intensity of an industry.
5.2 Pure Learning-by-Exporting Model
To eliminate self-selection mechanism I will equalize the fixed cost of entering both domestic
and export market, and make the extra starting-up exporting cost K to be 0. This way the
threshold to the export-market is removed, and the firms can enter the export market under
the same condition as the domestic one. To be specific, I have ηij = κ0, ∀i, j. Under this
circumstance both high export prices and learning-by-exporting drives firms’s incentive to
export.
Next I recalibrate the pure learning-by-exporting model to replicate the stylized facts.
It is found that as long as we have a higher expor price, non-exporters will always jump
immediately. Even with no lower innovation cost and therefore no learning effect, an exporter
status is highly desirable due to the higher export profit. To replicate the empirical targets,
I need to allow a lower export price. Specifically, P1 should be 86% of P0. At the same time,
the learning effect must be shrunk significantly; specifically domestic innovation cost should
be only 0.3% higher than export innovation cost.
Based on this renewed parameterization, the pure learning-by-exporting can now success-
fully replicate the relative export supply. However, even with the same supply ratio, there
are much more exporters under this extreme case. The exporter percentage will increase to
80%, while in the data it is only 25%. Therefore, compared with the benchmark model, the
export intensity will be greatly encouraged in terms of a soaring number of exporters on the
market. Without self-selection mechanism and higher export price, the positive correlation
between exports and production can be partially explained with an unconvincingly high ex-
porter percentage. As indicated by the data, even after decades of globalization, exporting
behavior is not considered common among producers nowadays.
Figure 17 in Appendix C depicts the simulated growth path of a firm beginning from q
= 0.1. It now takes 2 periods for a non-exporter to jump. Born as a non-exporter, a firm
will wait for only one period before it starts the exporting. Even with a lower export profit,
12
Table 2: Pure Self-Selection vs. Pure Learning-by-Exporting
Description Benchmark Pure SS Mech. Pure LBE Mech.
Time to Jump1 9 9 2
Critical q∗ 2.79 2.73 0.35
Steady-State Firm-Level Results:
Innovation 0.72 0.66 0.60
Exporting Ratio 0.32 0.34 0.17
Production 3.59 3.30 2.99
Data Replication Results:
Relative Exporting Supply QS1 /Q
S0 0.19 0.15 0.19
Exporter Percentage 25.03 25.00 80.00
1: It indicates how many periods for a firm starting with q = 0.1 to switch from anon-exporter to an exporter.
being an exporter is still very desirable because of a lower innovation cost. The innovation
investment is the highest at the very first period, then gradually decreases due to the increasing
marginal cost. All the steady-state values on the firm level are much lower compared with
the benchmark model; the innovation investment and production both decrease by 17%, and
the export ratio decreases by 47%. On the market level, the relative export ratio can be
replicated, but the exporter percentage is as much as 3.2 times in the data. Therefore, the
self-selection mechanism can slow down the export-status switching process, and keep the
industry’s export intensity under control.
Table 2 summarizes the results. The steady-state innovation, export ratio and produc-
tion are on firm-level; the steady-state relative export ratio and exporter percentage are on
aggregate-level. Generally speaking, self-selection mechanism works in a sense of aggregate-
level adjustment, and allows more productive firms into the export market. By establishing
a export-entry threshold, the self-selection mechanism allows time for a non-exporter to grow
until it starts to export, thus separate the domestic market from the export market. On
the other hand, the learning-by-exporting mechanism works in the sense of individual-level
development; after a firm starts to export, it benefits from a lower innovation cost — as an
important motivation to jump. Without either mechanism the model fails to replicate the
13
data adequately enough. Pure self-selection can only replicate the exporter percentage but
with a lower relative export supply; pure learning-by-exporting can only replicate the relative
export supply but with a much higher exporter percentage. Only with both mechanisms can
the model adequately explain both stylized facts.
6 Trade Liberalization
In this section I will make counterfactual analysis of trade liberalization. Assume now an
exogenous trade liberalization happens, and the firms face a lower transportation cost. To be
specific, recall that in the benchmark model, the iceberg transportation cost τ was normalized
to be 1. Now τ is considered to be variable. Under a trade liberalization, I assume that
this transportation cost decreases by 10%. As a result, the real export price P1/τ increases
exogenously from 2 to 2.2. This liberalization can happen abruptly within one period, or
can gradually happen over several periods. I will analyze the changes of firm-level dynamics
under different scenarios in details. I then make comparison between the counterfactual and
the previous benchmark results to determine the influence of different liberalizations.
6.1 Mature Industry
The first counterfactual case is a liberalization after the aggregate industry has reached its
steady state. In the benchmark model, at the 39th period the industry will be stabilized. Now
assume that a industry is mature enough to have already achieved stabilization, then it is
disturbed by a decrease of the transportation cost. As a result, the real export price is raised
by 10%. Because of this change, the firms will make different decisions on their innovation
and export ratio. This will lead to changes in aggregate supplies and exporter percentage
from time to time. Panel A of Table 3 shows these changes over different periods.
Difference cases in Panel A explain different liberalization scenarios: first, when the real
export price rises abruptly within one period; second, P1/τ rises gradually for 5 periods, each
period increases by 0.04; third, P1/τ rises gradually for 10 periods, each period increases
by 0.02. Now compare these scenarios at the same period after the liberalization finishes.
Across different cases there are no changes in the exporter percentage over time, and only
slight difference in the relative export supply. In the middle run, after 25 periods of the
liberalization, a longer liberalization brings a lower QS1 /QS0 , meaning that the less products
are supplied to the export market. However this decrease of relative export supply is only
around 0.02% between different scenarios. In the long run, there are no significant differences
of exporter percentage across these scenarios either. If trade liberalization happens to mature
industry which is already stabilized, then whether the change happens abruptly or gradually
won’t matter; the export intensity of the industry will increase slowly period by period. To
14
Table 3: Trade Liberalization Counterfactual Dynamics
Panel A Trade Liberalization for a Mature Industry (after the 39th period)
Case 1a Case 1b Case 1c
Abrupt Change Gradual 5 Periods Gradual 10 Periods
QS1 /Q
S0 Exporter % QS
1 /QS0 Exporter % QS
1 /QS0 Exporter %
Short Run (1 period later) 0.191 26.15 0.191 26.22 0.192 26.26
Middle Run (5 periods later) 0.192 26.22 0.192 26.26 0.192 26.28
Mid-Long Run (25 periods later) 0.292 30.29 0.292 30.30 0.292 30.30
Long Run (30 periods later) 0.299 31.30 0.299 31.30 0.299 31.30
Panel B Trade Liberalization for a Young Industry (the 8th period)
Case 2a Case 2b Case 2c
Abrupt Change Gradual 5 Periods Gradual 10 Periods
QS1 /Q
S0 Exporter % QS
1 /QS0 Exporter % QS
1 /QS0 Exporter %
Short Run (1 period later) 0.101 4.087 0.138 14.21 0.135 18.69
Middle Run (5 periods later) 0.145 15.35 0.175 19.77 0.180 22.03
Mid-Long Run (25 periods later) 0.207 26.97 0.201 26.05 0.194 24.72
Long Run (30 periods later) 0.209 27.44 0.203 26.30 0.193 24.83
Panel C Trade Liberalization for a Young Industry (the 5th period)
Case 3a Case 3b Case 3c
Abrupt Change Gradual 5 Periods Gradual 10 Periods
QS1 /Q
S0 Exporter % QS
1 /QS0 Exporter % QS
1 /QS0 Exporter %
Short Run (1 period later) 0.083 6.522 0.042 1.565 0.028 1.155
Middle Run (5 periods later) 0.161 14.55 0.111 10.92 0.085 7.304
Mid-Long Run (25 periods later) 0.311 33.47 0.170 20.80 0.125 18.69
Long Run (30 periods later) 0.313 34.24 0.172 21.19 0.126 18.86
15
be specific, after 30 periods the relative export supply will grow by 57.37%, and the exporter
percentage will grow by 25.20%.
6.2 Young Industry
Now assume a trade liberalization happens before the aggregate industry reaches its initial
steady state. Specifically, P1/τ rises when the industry is still at its initial stage of growth.
I then look into how the dynamics of the industry differs across scenarios with different
liberalization patterns.
I consider two cases: first, liberalization upon a young industry, which is at its 8th period of
growth; second, liberalization upon a even younger industry, at its 5th period. The dynamics
of supply ratio and exporter percentage over time are shown in Panel B and C of Table 3.
In Panel B when the trade liberalization happens, only partial firms in the industry have
started exporting. After the abrupt liberalization takes place, immediately the the exporter
percentage will rise. Unlike the previous results in section 6.1, the relative export supply will
be lower compared with the benchmark results at the same period. As time goes by, the
relative export supply starts to rise and 30 periods later, it ends up at a higher level (9.30%
higher) than the contemporaneous benchmark result. The exporter percentage, on the other
hand, will consistently increase. At the 30th period after the liberalization, the exporter
percentage ends up being 8.46% higher than the benchmark level.
If the liberalization takes 5 periods to gradually finish, right after the change happens the
relative export supply QS1 /QS0 will be 1.57% lower than the benchmark result, however the
exporter percentage will not change. As time goes by, both the relative export supply and
the exporter percentage will increase, and 30 periods later the export supply ratio is 6.18%
higher than the contemporaneous benchmark result. Interestingly enough, at this moment
the exporter percentage happens to be the same as the benchmark result.
If the liberalization takes 10 periods to finish, things are again different. Immediately
both relative export supply and exporter percentage will be lower than contemporaneous
benchmark results. As time goes by, both of them will rise. However not both of them can be
as high as the benchmark results even after a long while. To be specific, after 30 periods the
relative export supply will be only 0.81% higher than the contemporaneous benchmark result,
while the exporter percentage will be 1.86% lower. To sum up, if trade liberalization happens
to a young and unstabilized industry, it does not necessarily increase the export intensity of
the industry. The result depends on how the change takes place; if it takes place abruptly
then both relative export supply and exporter percentage will rise. If it takes as long as 10
periods to complete, then the exporter percentage will even be lower than a no-liberalization
level for a very long time.
In Panel C, the trade liberalization happens to a even younger industry at its 5th period,
16
when no exporters appear yet. If the change takes place abruptly, then immediately there
will be exporters. Again, the relative export supply as well as the exporter percentage will
be higher than the contemporaneous benchmark results. As time goes by the relative export
supply gradually rises. After 30 periods of the liberalization, the relative export supply is as
1.64 times as much in the benchmark level, while the exporter percentage is 1.37 times as
much as in the benchmark level. However, if the change takes place gradually, there will be
much less exporters right after the liberalization finishes. As time goes by the relative export
supply and the exporter percentage will both increase. After 30 periods, neither of them
is as high as the contemporaneous benchmark results. To be specific, if the liberalization
lasts for 5 periods, in the long run the relative export supply will be 10.29% lower, and the
exporter percentage will be 16.25% lower. If the liberalization lasts for 10 periods, in the long
run the relative export supply will be 33.68% lower while the exporter percentage 25.24%
lower. Therefore surprisingly, if the liberalization happens to an industry that is so young
that no exporters have appeared yet, the longer the trade liberalization lasts, the lower export
intensity the industry will have for a long time.
To sum up, the dynamics of the export intensity varies across different exogenous trade
liberalization patterns. First, if the liberalization happens to mature industry which has
already reached its steady state, across various liberalization patterns no significant difference
will happen. Second, if the liberalization happens to a young industry with partial firms
as exporters, the longer the liberalization takes, the higher relative export supply will be.
In other words, the relative export supply is lower because of a slower trade liberalization.
Third, if the liberalization happens to a even younger industry with no exporters yet, the
relative export supply and the exporter percentage will both instantly increase at once if the
change is abrupt. But if the change takes 5 or 10 periods for the change to gradually finish,
the relative export supply and the exporter percentage will both be lower compared with
contemporaneous benchmark results, and the export intensity of an industry will be reduced
significantly for a long time. Finally, through comparing between different cases, we can find
that the abrupt liberalization takes effect much faster for a young industry than a mature
industry.
Therefore, to determine whether an exogenous trade liberalization can encourage the ex-
porting behavior in an industry, we need to consider when the change happens, and how long
it lasts. If policymakers want to use a trade liberalization to effectively encourage the export-
ing behavior in an industry in terms of export supply and the number of exporters, there are
important things they should keep in mind: either wait until the industry is stabilized, and
the liberalization pattern won’t matter. If the industry is still very young, then implement the
liberalization as fast as possible; otherwise the export intensity will probably be discouraged
instead for a long period. Under this circumstance, the industry’s export intensity will be
17
encouraged much faster than in a mature industry.
Figures 20, 21 and 22 in Appendix C depict the distribution of firms’ innovation, export
and production after 40 periods. Each figure shows the comparison among different distribu-
tions under a specific trade liberalization pattern. Liberalization after the steady state brings
no change of the distribution across figures under different patterns. If the liberalization takes
place even before any firm starts to export, variations are the most significant when it comes
to the distribution of production and exporting ratio. Specifically, the earlier and the slow-
er the liberalization happens, more firms will have lower production, and there will be less
exporters on the industry.
7 Conclusion
In this paper, I have developed a dynamic model of a firm’s joint decisions to export and
innovate, allowing both decisions to affect the firm’s production in accordance with self-
selection and learning-by-exporting theories of firm-level export and production dynamics.
The model is extended to allow for an industry containing many firms that are heterogenous
with respect to age, export status, and production.
I find that both self-selection and learning-by-exporting critically influence the dynamic
evolution of a firm, an thus the dynamic evolution of an industry. On the one hand, models
that allow only for self-selection can replicate the percentage of firms that export, but predict a
much lower volume of exports than is observed. On the other hand, models that allow only for
learning-by-exporting can replicate the relative volume of exports, but predict unrealistically
high percentage of firms involved in exporting. To adequately capture the stylized facts
embodied in Chilean manufacturing from 2005 to 2007, both mechanisms must be posited to
be effective.
In a counterfactual policy analysis, I find that the effects of trade liberalization depend
critically on the age of the industry. One the one hand, liberalization trade for a mature in-
dustry encourages exports. However if the trade liberalization happens to a young industry at
its initial stage of production evolvement, and it takes long enough to finish the liberalization,
the export intensity of the industry will be reduced.
Also, there remain may potentially interesting extension of the work presented in this
paper. For example, the model’s predictions regarding the impacts of trade liberalization
can be carried out. Another theoretical extension would be to allow firms to shut down
temporarily, rather than shutting down forever. Finally, one could extend the model by
explicitly incorporating a production function and factor pricing. This would allow us to
study the influence of wages on firm- and industry-level dynamics and the impacts of labor
immigration and wage policies.
18
References
[1] Alvarez, R., and Lopez, R. Exporting and firm performance: Evidence from chilean
plants. Canadian Journal of Economics 38(4) (2005), 1384–1400.
[2] Amiti, M., and Konings, J. Trade liberalization, intermediate inputs, and productiv-
ity: Evidence from indonesia. American Economic Review 97(5) (2007), 1612–1638.
[3] Atkeson, A., and Burstein, A. Innovation, firm dynamics, and international trade.
Journal of Political Economy 118(3) (2010), 433–484.
[4] Aw, B., Roberts, M., and Winston, T. Export market participation, investments
in r&d and worker training, and the evolution of firm productivity. The World Economy
30(1) (2007), 83–104.
[5] Aw, B., Roberts, M., and Xu, Y. R&d investments, exporting, and the evolution of
firm productivity. AER Papers and Proceedings 98(2) (2008), 451–456.
[6] Baldwin, R., and Gu, W. Export market participation and productivity performance
in canadian manufacturing. Canadian Journal of Economics 36(3) (2004), 634–657.
[7] Ben-David, D. Equalizing exchange: Trade liberalization and income convergence.
Quarterly Journal of Economics 108(3) (1993), 653–679.
[8] Bernard, A., Jensen, J., and Kortum, S. Plants and productivity in international
trade. American Economic Review 93(4) (2003), 1268–1290.
[9] Bernard, A., Jensen, J., and Schott, P. Trade costs, firms and productivity.
Journal of Monetary Economics 53(5) (2006), 917–937.
[10] Blalock, G., and Gertler, P. Learning from exporting revisited in a less developed
setting. Journal of Development Economics 75(2) (2004), 397–416.
[11] Bustos, P. The impact of trade on technology and skill upgrading evidence from
argentina, 2005. CREI and Universitat Pompeu Fabra, Working Paper.
[12] Bustos, P. Multilateral trade liberalization, exports and technology upgrading: Ev-
idence on the impact of mercosur on argentinean firms. American Economic Review
101(1) (2011), 304–340.
[13] Clerides, S., Lach, S., and Tybout, J. Is learning by exporting important? micro-
dynamic evidence from columbia, mexico and morocco. Quarterly Journal of Economics
113(3) (1998), 903–947.
19
[14] Costantini, J., and Melitz, M. The Dynamics of Firm-level Adjustment to Trade
Liberalization. Cambridge: Harvard University Press, 2007.
[15] Hallward-Driemeier, M., Iarossi, G., and Sokoloff, K. Exports and manufac-
turing productivity in east asia: A comparative analysis with firm-level data. NBER
Working paper No. 8894 (2002).
[16] Imbruno, M. International trade and firm productivity within the italian manufacturing
sector: Self-selection or learning-by-exporting?, 2008.
[17] Judd, K. Numerical Methods in Economics. MIT press, 2004.
[18] Lileeva, A., and Trefler, D. Improved access to foreign markets raises plant-level
productivity...for some plants. Quarterly Journal of Economics 125(3) (2010), 1051–1099.
[19] Marin, D. Is the export-led growth hypothesis valid for industrialized countries? The
Review of Economics and Statistics 74(4) (1992), 678–688.
[20] Melitz, M. The impact of trade on intra-industry reallocations and aggregate industry
productivity. Econometrica 71(6) (2003), 1695–1725.
[21] Miranda, M., and Fackler, P. Applied Computational Economics and Finance. MIT
press, 2004.
[22] Olley, G., and Pakes, A. The dynamics of productivity in the telecommunications
equipment industry. Econometrica 64(6) (1996), 1263–1297.
[23] Pavcnik, N. Trade liberalization, exit, and productivity improvements: Evidence from
chilean plants, 2000. NBER working paper 7852.
[24] Verhoogen, E. Trade, quality upgrading and wage inequality in the mexican man-
ufacturing sector: Theory and evidence from an exchange rate shock, 2004. Columbia
University, Working Paper.
[25] Wagner, J. Exports and productivity: A survey of the evidence from firm-level data.
The World Economy 30(1) (2007), 60–82.
[26] Wu, R. Reconsider learning by exporting hypothesis from innovation: An empirical
study of us industries, 2011. The Ohio State University, Working Paper.
[27] Wu, R. Empirically revisiting the learning-by-exporting theory using data on chilean
manufacturing plants, 2012. The Ohio State University, Working Paper.
20
A Policy Analysis
In this section I will perform a policy analysis of the model. That is to analyze the influence
of relevant parameters on the computational results, given potential policies that vary these
parameters. The analysis is targeting on these seven variables: the period for a initially non-
exporter starting with q = 0.1 to jump; the firm-level steady-state innovation investment,
export ratio, and production; the aggregate-level steady-state aggregate domestic supply,
export supply and exporter percentage.
A.1 Export Price P1
Previously in Section 5 I have already done counterfactual analysis based on change of P1.
Here are more details of how it changes the economy. Keep the domestic price P0 to be equal
to 1, and all the other parameters to be the same as the benchmark model; now suppose the
government is imposing an export subsidy or tax on the exporters so that the real export
price received is varying. Figure 6 reports all the relevant difference P1 can make. Panel (a)
depicts how an individual firm starting with q = 0.1 will be affected by changing P1, while
panel (b) how the entire market will be affected.
Figure 6: Simulation Result of a Representative Firm and a Market –Sensitivity analysis of Export Price P1
If P1 is less than 1.96, then a non-exporter will never become an exporter since there is not
enough profitability of exporting. In this case the steady state innovation investment, export
ratio as well as production will keep constant. If P1 equals 1.96, a non-exporter beginning
with q = 0 will be able to jump. If P1 increases further, a non-exporter needs fewer and
21
fewer periods to jump. Intuitively speaking, as the profitability of exports increases, it is
more attractive and more affordable for a firm to start exporting sooner than later. Thus the
steady-state export ratio will also rise as P1 rises; an exporter will eventually choose to export
more out of its entire production. Meanwhile, the steady-state innovation investment will be
much higher and start to rise once a firm is capable to jump, and so will the steady-state
production.
As for the steady-state of an entire market, if P1 < 1.96, there will be neither exporters
nor export supply; the steady-state aggregate production is the aggregate domestic supply,
which is fixed. As the export price exceeds 1.96, the steady-state domestic supply start to
drop; the export supply rises, and so does the exporter percentage. The more subsidy the
government gives to the exporters, the more exporters there will be on the market.
Figure 7: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Innovation Cost Parameter - Exporter c1
A.2 Innovation Cost Parameter - Exporter c1
Now let’s look into how c1 — the innovation cost for the exporters — influences the steady-
state values. Figure 7 shows the results. Suppose the government impose an additional tax on
the exporters so that their innovation become more expensive (c1 rises). It costs the exporters
more in innovation. In this case, it takes longer for a non-exporter to start exporting. The
steady-state innovation will decrease due to the higher production cost; so will the steady-
state production level. Yet the steady-state exporting ratio will rise, even though it means
more costly investment in the innovation. As for an entire market with heterogeneous firms,
as c1 increases there will be a decrease in both export supply and exporter percentage, due
22
to the rising production cost. The steady-state domestic supply will also decline slightly.
A.3 Innovation Cost Parameter - Non-Exporters c0
c0 is the innovation cost for the non-exporters. Figure 8 shows how this parameter influences
the results. Now suppose the government impose a tax on non-exporters’ innovation and c0
rises. It means it costs non-exporters more to invest in innovation. Because of the higher
cost, it makes non-exporters grow slower, and takes longer to start exporting. But as long
as the firms can eventually become exporters, the steady-state innovation, exporting ratio as
well as the steady-state production will not be affected. Varying between 0.5 and 3, c0 allows
this jump to happen. Yet as for the market, the total domestic and export supply will drop
due to the higher cost, and so will the exporter percentage.
Figure 8: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Innovation Cost Parameter - Non-Exporter c0
A.4 Production Cost Parameter - Exporting Products d1
Figure 9 shows the influence of d1, the production cost for exporting products. As it increases,
it costs more to produce the exporting products. The more a firm exports, the more it needs
to pay for production. Under the benchmark parameter setting, non-exporters will be able
to jump eventually when d1 < 0.86. As d1 rises, non-exporters will need more time to start
exporting. Also, the firm’s steady-state export ratio will decrease as d1 increases because of
the higher cost. Also, the firm’s steady-state innovation will decrease, leading us to a lower
production. As d1 rises, the exporters’ export behavior will be hindered and the total export
23
supply as well as exporter percentage will both drop. The total domestic supply will rise, on
the other hand.
Figure 9: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Production Cost Parameter - Exporter d1
Once d1 > 0.86, a non-exporter can never make the jump, and its steady-state innovation
and production will be at a much lower level. Also, in this case, its steady-state innovation
and production cannot be changed by d1 since d1 does not affect the firm’s decision anymore.
The market will also be unaffected by d1 via a constant domestic supply and no exporting
activities.
A.5 Production Cost Parameter - Domestic Products d0
d0 is the production cost for domestic products. Every firm, except those 100% exporters, will
be influenced by it. As described in Figure 10, as d0 increases, the more costly a firm’s domestic
production becomes. Non-exporters will be never jump when d0 < 0.16 since producing purely
domestically is more profitable. In this case as d0 rises, non-exporters will decrease their
steady-state innovation due to the higher cost, resulting in a lower production. The total
domestic supply in the market will slightly fall.
Once d0 > 0.16, a non-exporter will now start to export, since purely domestically pro-
duction is becoming more expensive. Interestingly, the firms’ steady-state innovation and
production will start from a higher level, then decrease as d0 increases. Also, its steady-state
export ratio is rising as it costs more and more to produce domestic products, and it takes
fewer periods for a non-exporter to jump. As a result, total export supply as well as the
exporter percentage will both increase.
24
Figure 10: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Production Cost Parameter - Non-Exporter d0
A.6 Production Depreciation Rate γ
Figure 11 shows the analysis based on γ, the depreciation rate of production from today to
tomorrow. Suppose there is an economic recession which increases γ and creates an unfavor-
able environment for the firms to grow. Consequently q will grow slower. Specifically, as γ
increases below 0.23, the periods it takes for a non-exporter to jump to be an exporter will
increase, since it will take more time for a firm’s production to evolve.
Figure 11: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Production Depreciation Rate γ
25
Furthermore, the rising γ will increase the steady-state innovation and export ratio. A
firm is urged to innovate and export more, so as to keep its production from falling too fast.
However because of the rising depreciation rate, its steady-state production is still declining
even though the firm increases its innovation and exporting effort. As a result, there are fewer
and fewer exporters on the market.
Once γ exceeds 0.23, non-exporters can never make the jump and there will be no exporters
eventually. The firm’s steady-state innovation and production will both sharply fall to a much
lower level. As γ keeps rising, the firm will increase its innovation effort even though now
it can no longer learn from exporting. But still, its steady-state production keeps falling
due to the higher depreciation rate. The total domestic supply always keeps decreasing as γ
increases.
A.7 Fixed Cost - Exporter κ1 and Exporting Start-Up Cost K
κ1 is the fixed cost a firm needs to pay once it enters the exporting market. Assume the
foreign tariff rises, so does the cost. In Figure 12, as κ1 increases, a domestic firm has to
pay more to enter the exporting market. Therefore, when κ1 < 2.03 a non-exporter is able
to jump eventually. As long as the jump happens, as κ1 rises it will take more time for a
non-exporter to grow until it reaches the exporting threshold. The more a firm needs to pay
for its entry to the exporting market, the longer it must wait until it can afford this cost. The
steady-state values of innovation and production will not change.
Figure 12: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Fixed Cost - Exporter κ1
Once κ1 > 2.03, the fixed exporting cost will be so high that a non-exporter can never
26
make the jump, and there will be no exporters on the market in the steady-state. Once a
non-exporting firm can never afford the switch, its steady-state innovation will drop sharply;
as a non-exporter, without any learning from exporting, its production will also drop. Also,
if a firm always stays as a non-exporter, its steady-state innovation and production will not
change as κ1 changes.
The analysis of the exporting start-up cost K has the similar results, which are depicted
in Figure 13. If K < 1.40, non-exporters can make the jump. As K increases, it costs more
for a non-exporter to enter the exporting market, therefore it takes longer for it to jump.
There will be less exporters on the market, and the domestic supply will increase. Firms’
steady-state innovation, exporting ratio and production will keep the same if non-exporters
can make the jump, and keep the same at another lower level if no jumps can ever happen.
To sum up, if the exporting entry threshold changes, the steady-states of individual firms
will not be affected unless this change forbids exporting behavior. Only the steady states of
aggregate economy will be different.
Figure 13: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Exporting Start-Up Cost K
A.8 Innovation Cost Rate β
Let’s look at how β affects the steady-state results of the model. β is a firm’s innovation cost
rate. Notice that the steady-state innovation is always below one. The higher β becomes, the
less it costs a firm, no matter a non-exporter or an exporter, to pay for the same innovation
investment. In Figure 14, as I increase β from 6 to 12, it costs less to innovate. The periods
it takes for a new-born firm to jump will decrease, thus makes the firm to innovate more and
27
grow faster. Meanwhile, increasing β can increase a firm’s exporting ratio, because the firm
needs to resort to exports to keep growing. Thus steady-state production will increase as β
rises.
Figure 14: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Innovation Cost Rate β
Assume the government offers subsidies to firms’ innovation investment, and this policy
cut the innovation cost greatly. In the aggregate market, because higher β and lower cost, both
domestic and exporting supply will increase, and the exporter percentage will also increase
fast.
A.9 Death Probability µ
µ is the probability of a firm being forced to exit the market. In other words, each period a
firm will have a chance of µ to stop production and discontinue its existence on the market. As
this probability varies, the steady-state policies and production of a firm will not be affected;
individual decision of how much to innovate and export will be not affected by the death rate.
As in Figure 15, only the aggregate steady-states of an entire economy will be different
because of the change of µ. Assume the government decides to increase the liquidity of the
market, so that the frequency of firms’ in-flow and out-flow increase. Thus µ goes higher, and
it is more likely for a firm to be forced out of the market; each period more firms die and
more new firms enter. What’s more, under the benchmark parameter setting Q < q1, which
means the new firms are all born as non-exporters. Thus the less the total supply to both
domestic and foreign market will become, and the less exporters there will be.
28
Figure 15: Simulation Result of a Representative Firm and a Market –Sensitivity Tests of Death Probability µ
A.10 Summary & Discussion
There is a major finding based on the above sensitivity tests. The steady-state values of
the production all experience a sharp increase once a firm seizes its opportunity to become
an exporter. Compared with an eternal non-exporter, a firm that ends up as an exporter
will have a significantly higher production. Correspondingly, the steady-state innovation and
exporting ratio will both be significantly higher. The virtuous circle will promote the growth
of an exporter much further. Again, this finding supports the learning-by-exporting theory
— the exporters can experience faster production growth compared with the non-exporters.
Furthermore, a major finding based on the benchmark model is that once an exporter,
always an exporter. However in the real life, as what the data indicate, there are almost as
many firms switching forward to be exporters as those switching back to be non-exporters. To
explained the firms’ withdrawal from the exporting market, changes should occur to relevant
parameters, e.g. exporting price P1, or production depreciation rate γ. In the benchmark
model all these parameters are given, therefore the firms will always keep their exporting
status. But if a new trade agreement is made, and some industries suddenly suffers from a
declining exporting price; or during a recession the production depreciation rate is soaring
high. Then the withdrawal from exporting will happen and certain firms will switch back and
never become exporters again.
29
B Table
Table 4: Descriptive Statistics, 2005 to 2007
Exporters of the year Non-Exporters of the year
Number Percentage Number Percentage
A. TOTAL NUMBER OF PLANTS 1,624 100.00 4,990 100.00
B. CAPITAL PROPORTION
Foreign Capital Proportion = 0 1,321 81.34 4,872 97.64
Foreign Capital Proportion ∈ (0 50%) 49 3.02 42 0.84
Foreign Capital Proportion ∈ [50% 100%) 86 5.30 39 0.78
Foreign Capital Proportion = 100% 168 10.34 37 0.74
C. Technical Assistance
Foreign Tech Assistance Ratio = 0 1,321 81.34 4,872 97.64
Foreign Tech Assistance Ratio ∈ (0 5.0e-6) 167 10.28 43 0.86
Foreign Tech Assistance Ratio ∈ [5.0e-6 1.0e-5] 47 2.89 11 0.22
Foreign Tech Assistance Ratio > 1.0e-5 89 5.48 64 1.28
D. SIZE
Small (10-49 Workers) 669 41.19 4,377 87.71
Median (50-149 Workers) 895 55.11 595 11.92
Large (≥ 150 Workers) 60 3.69 18 0.36
E. VALUE ADDED RATIO
< 0 12 0.74 37 0.74
∈ [0 0.5) 1,088 67.00 3,095 62.02
≥ 0.5 524 32.27 1,858 37.23
F. INNOVATION ON ASSETS
Innovation on Buildings > 0 122 7.51 163 3.27
Innovation on Vehicles > 0 15 0.92 30 0.60
Innovation on Machines > 0 101 6.22 144 2.89
Innovation on Land > 0 7 0.43 14 0.28
No Innovation 1,436 88.42 4,711 94.41
G. Productivity
a ∈ [0 0.5) 56 3.45 1,252 25.09
a ∈ [0.5 1.5) 391 24.07 2,149 43.07
a ∈ [1.5 3) 1145 70.50 1,563 31.32
a ≥ 3 32 1.97 26 0.52
Source: ENIA (Annual National Industry Survey), National Institute of Statistics of Chile.
30
C Figures
Figure 16: Simulation Results of the Growth Path of a Firm Starting with q = 0.1
Figure 17: Simulation Results of the Growth Path of a Firm Starting with q = 0.1
31
Figure 18: Density of Productivity of New-Born Firms
Figure 19: Density of Productivity of Exporters vs. Non-exporters
32
Figure 20: Cumulative Distribution of Firm Production 40 Periods Later:Abrupt Trade Liberalization within 1 Period
33
Figure 21: Cumulative Distribution of Firm Production 40 Periods Later:Gradual Trade Liberalization for 5 Periods
34
Figure 22: Cumulative Distribution of Firm Production 40 Periods Later:Gradual Trade Liberalization for 10 Periods
35