Factors driving the firms decision to export. Firm-level evidence from Poland.

Munich Personal RePEc Archive

Factors driving the firms decision to

export. Firm-level evidence from Poland.

Hagemejer, Jan

National Bank of Poland, Warsaw University, Poland

6 June 2007

Online at https://mpra.ub.uni-muenchen.de/17717/

MPRA Paper No. 17717, posted 08 Oct 2009 13:52 UTC

Firm-level evidene from Poland.

Jan Hagemejer

National Bank of Poland

June 6, 2007

Abstrat

ThemodelbyMelitz(2003)preditsthatifrmsdierintheirprodutivity(TFP)

and there exists a xed osts of entry to export markets, rms begin exporting if

produtivity exeeds a ertain threshold value. Produtivity is thus a ruial fator

behind rms' export market partiipatio n. To verify this, I estimate a simple probit

model of the rms deision to export, based on the Polish manufaturing rm-level

data. EstimationofprodutivityofindividualrmsistroublesomeasthestandardOLS

methodproduesbiasedestimatesduetotheendogeneityoffatorhoie. Iuseamulti-

stagesemi-parametriapproah, asproposedbyOlleyandPakes(1996)ontrolling for

endogeneity and the bias aused by rms exiting and entering the sample during the

periodunderonsideration. BesidesdeterminingthesignianeoftheTFP oeient

intheprobitregression,Iexaminethepathsofprodutivityofrmsenteringtheexport

market and make an attemptto identify the potential learning-by-expor ting eets.

Keywords: produtivity, exports,rm-level data

JEL lassiation: F10 F14D21L60

Empirialliteratureoninternationaltradeseemstograduallydriftaway fromtheonept of

symmetri rms within an industry. Analysis of rm level data indiates, that there exists

not onlya great deal of heterogeneity among rm, but there are also signiantdierenes

in rm behavior. One of the topis that has reently attrated a lot of attention of both

the empirial and theoretial literature is the fat that onlya fration of rms in any given

industry deidesto exports while the rest isonly supplyingto domesti market.

Theoretial literature provides the following explanation of this phenomenon. Initiation

of exports requires bearing some xed and sunk osts of entry and the rm has to generate

a suient level of prots to make sure that it an aord entry intoexport market. Thus,

more eetive rms export while the less eetive rms are below the required eieny

threshold and deide to stay away from the foreign market. Besides the above mehanism,

there is another intuitive hannel of interation between exports and produtivity. Firms

engaginginontatswithothermarketsanbenetfromexperieneof foreignrmsand use

theseknowledgeindomestimarkets. Moreover, rmsompetingintheforeign marketmay

try harderin terms of quality of their produts whih inturn alsoaets homeonsumers.

This artileis an attempttoexplain the determinants of export deisionof Polish rms

in the period 1997-2004. The fators that has been taken into onsideration are rm pro-

dutivity and rm size and other rm harateristis. The regression analysis inludes also

suhsetoralfatorsasexportpenetration,industryonentrationandthe existeneofteh-

nial barrierstotrade. Anattempthas been madeto verify the ausality diretion between

produtivity and exporting.

The artilehas a followingstruture. In the rst setion I review the relevant empirial

and theoretial literature related to rm heterogeneity and international trade. Seond

setion presents the theoretial bakground behind the estimation equation. A detailed

desription of inluded variables and data used is ontained in setion three. Setion four

follows with the estimationresults together with sensitivity analysis and Granger ausality

1 Literature review

Traditionaltrade theory is based onanassumption of onstant returns tosale and perfet

ompetition. Thanks to these assumptions, all onlusions are formulated on the industry

levelandindividualrmbehaviorisregardedasalmostnotimportantasitdoesnothaveany

impaton the industry situation. This theory annotexplain many issues that haraterize

moderninternationaltrade,suhasintra-industrytrade. The diretionand volume oftrade

is determined either by omparative advantage (the Riardian framework) or by relative

endowment of fators of prodution(Heksher-Ohlin model).

The soallednewtrade theoryassoiatedusually withsuhnamesasKrugman orHelp-

man seems topartially solve the problems. In the Krugman (1980)model, monopolistially

ompetitive rms exports their produts thanks to onsumers haraterized by a love-for-

varietyutilityfuntion(gettingahigherutilitylevelthankstoextravarietiesimported). The

KrugmanandHelpman(1985)modelextendstheanalysisbyelementsoftheHeksher-Ohlin

model, allowing for the impat of relative fator endowments on the diretion and volume

of trade. These models, while learly being probably the most important ontributions to

the international trade literature in the seond half of the XX entury, are based on the

representative rm assumption - all rms in an industry are idential and make idential

deisions. If one of them deidesto export, all othersfollow.

Inspetion of Polish manufaturing rm-leveldata in the period of 1997-2004 (Table 1)

showsthatnotallrmsexport. Dependingontheriterionusedtolassifyrmsasexporters,

the perentage of rms that export is between 61 and 76 perent in 2004. Moreover, the

frationof exporting rmsisvisibly hangingintime- inthe 1997-1999period,the fration

ofexporting rmswasvisiblylowerthanin2004. It isworthnotingthatthe sampleofrms

usedtoprepare table1ontainsonlydataonlargermsthatemployover50people. Similar

Shareof exporters

year

X >

⁰

_{P KB} ^X > 0 . 01 _{P KB} ^X > 0 . 025

1997 71,44% 58,31% 51,80%

1998 70,36% 58,13% 51,95%

1999 69,78% 56,54% 50,10%

2000 71,00% 58,53% 52,37%

2001 72,54% 60,10% 54,04%

2002 70,70% 60,31% 53,82%

2003 72,01% 62,68% 57,56%

2004 76,07% 67,04% 61,30%

First olumn showsperentage ofall rms thathadpositive exports,

olumnstwo and three, perentageof rms where exportsto revenue ratios

were higherthan the given threshold.

alulationsfortheUnitedStates(Bernard,Eaton,JensenandKortum2003)revealsslightly

dierent distribution of rms. In 1992, only 21 perent of Amerian entreprises exported

theirprodutandtwothirdsofthemexportedlessthan10perentofthevalueoftotalsales.

Empirialresearh in otherountries alsoquestions the representative rm assumption.

The theoretialliteraturemodelingheterogeneity ofrmbehaviorisprobablythe fastest

growing branhof internationaltrade researh urrently. The mostimportantontributions

so far are without doubt the works by Melitz ((2003), with further extensions) or Bernard

et al. (2003). The Melitz model is in its struture slightly similar to the Krugman (1980)

model. The demand side is almostidential (onsumers are haraterized by a CES utility

funtion). The supply sideassumes, that every rm'sprodutivityisrevealedtoher (drawn

from an exogenous probability distribution) before the entry, exit or export deisions are

made. Entry into export market involves xed osts. Firm enters export markets if the

present value of doing so is exeeding the value of restriting supplies to the home market.

Melitz shows that rm willenter the export marketwhen its produtivity exeeds aertain

threshold value.

There are some important impliations of the Melitz model. First, rms, whose pro-

or exit the industry. Seond, trade liberalization indues some rms that did not export

before to start exporting. At the same time, with an inrease of the fator pries and a

shift of resoures towards exporting rms, the least exporting rms drop out of the market

(the produtivity threshold forthe rm presene inthe domesti marketshiftsupwards). It

means that trade liberalizationauses an inrease of average produtivity.

Bernard, Eaton, Jensen and Kortum (2003) build a model based on rm heterogeneity,

that assumes that rms ompete in a Bertrand fashion. The model assumes that inter-

national dierenes in osts are stemming from dierenes in fator pries. Similarly as in

Melitz,rmsareheterogeneousintermsoftheirmarginalost-onlysomeofthemself-selets

totheexportmarket. Themodelshowsthatexportingrmsgeneratehigherprots,aremore

produtiveandare largerthannon-exporters. The empirialveriationof themodel seems

to indiate good performane in the model in explaining the trends in Amerian rm-level

data.

0 2 4 6 8 0 2 4 6 8

.8 1 1.2

non−exporter

exporter

Frequency

Productivity

Figure1: produtivity of exporters and non-exporters

The literature itedabove postulates the existeneof aself-seletion mehanismof rms

into export market. The high-produtivity/low-ost rms deide to start exporting, while

theless eetivermsupply onlytodomestimarket. Doestherealityonrmthat? Figure

1shows thedistributionoftotalfatorprodutivity(TFP)forPolishrmsin2003 . Wean

see that the distributionof produtivityof exporters islearly shifted tothe rightrelatively

to non-exporters. Bernard, Eaton, Jensen oraz Kortum (2003) report 33perent advantage

ofexportersovernon-exportersintermsoflaborprodutivity. Therelativelylowerdierene

between exporters and non-exporters in the ase of Polish rms might stem out from the

fat that the Polish data ontains only large rms, and the export status isorrelated both

with produtivity and size of rms aswill beshown later.

Dierenes in eieny of rms with onnetion to export deision were analysed in

detail by Bernard and Jensen (1997) using a panel of 50-60 thousand rms. Produtivity

(measured by TFP, value added per worker et.) was regressed on rm level and setoral

ontrol variables and the exporting status. In all ases, the result suggest an advantage of

exportingrmsof12to24perentrelativetonon-exporters. Moreover, exportingrmswere

50-60 perent larger than others.

Another branh of literature is trying to explain the ausal relationship between the

produtivity level and exports. There exists a ommonbelief that export partiipation an

positivelyinueneprodutivity-theso-alledlearning-by-exportingeet. Atthesametime

the theoretialliterature postulates the self-seletion mehanismdesribed earlier. Clerides,

LakandTybout (1998)estimatethermexport partiipationequationtogetherwithaost

funtion, where, besides a set of ontrol variables, past export partiipated isinluded (the

study isonefor Moroo,Mexioand Columbia). Whilethe resultslearly indiatetheself-

seletionmehanism(fromprodutivitytoexporting), learning-by-exportingis presentonly

inseletedsetors. BothBernardandJensen(1999)andAw,ChenandRoberts(1997)arrive

atsimilaronlusions. Intheaseof theformer,astudybasedonAmerianrms data,past

export status is signiant for survival rates but does not have any impat on traditional

produtivity measures. The latter study, based on Taiwanese data, learning-by-exporting

eetsseemtobesigniantonlyforseleted setors. ArnoldandHussinger(2005)estimate

1

ThemethodofalulationofTFPisdesribedin detailin later

auses export but the opposite ausality is nonexistent.

Pavnik (2002) makes an attempt to explain the link between trade liberalisation and

produtivity, using Chilean data. The results show that both in setors where export pen-

etration is high and in export oriented setors trade liberalization auses an inrease in

produtivity. At the same time, Pavnik shows that rms of highest produtivity inrease

theirmarketshares aftertradeliberalization. This indiates arealloationof resouresfrom

less eetive to more eetive rms. Bernard, Jensen and Shott (2003) perform a similar

study fortheUnited Statesandshowthatthe inreaseinprodutivityisstrongerinsetors,

where trade osts dereased faster.

2 Theory and methodology

An empirial model of determinants of export deision of a rm is diretly motivated by

existing theoretial literature on heterogeneous rms, espeially the Melitz model (2003).

As was indiated earlier, a rm enters the foreign market when revenues from doing so

exeed the xed ost of entry. Similarly as in Arnold and Hussinger (2005) this ondition

an be formulated asfollows:

Export if:

R ^e _i,t − C _i,t ^e (Z _i,t ^e ) > 0,

⁽¹⁾

where

R

^{is revenue,}

C

^{- produtionand sales ost}

Z _it

^{- ost} determiningvariables. Sub- sript

e

^{indiates variables related to the export market. When there are xed (sunk) ost}

to export, the problem beomes dynami and an be summarized by the followingBellman

equation:

V _t =

^max

X t ∈{ 0 , 1 }

R ^e _t − C _t ^e (Z _t ^e ) − S(1 − X _t− 1 ) + δE (V t− 1 )

,

⁽²⁾

where

X _t

^isanexportpartiipationdummyvariable(subsripts

i

^weresuppressed)forperiod

t

^,

C _t

^{isprodutionost}

t

^{,notinludingthe ostof entrytoexport market}

S

^.

δ

^isadisount

fator. Equation(2)saysthat rmsmaketheexport deisionmaximisingurrentand future

prots fromthe presene inthe export market.

Exportdeisionismadeinthefollowingway. This formulationistaken fromArnold and

Hussinger (2005)(see also Roberts and Tybout 1997):

X _t =



 



 



1

^if

R ^e _t − C _t ^e (Z _t ^e ) + δ[E t (V t +1 |X t = 1) − E _t (V t +1 |X t = 0)] > 0 0

^otherwise

(3)

The rm will enter the export market if the prots from export in time

t

^inluding

the future expete d value of partiipating in the export market are positive.

E _t

^{stands for}

expete d value at time

t

^.

Vetor

Z _it

^{ontains the variables} determining the ost of arms. These mightbe either setor spei, time spei or rm-spei. Costs an be largely determined by rm-level

produtivity(TFP).Thisvariableisunobservablefortheresearher,howeveritisobservable

by the rm.

Lets assumethe standard Cobb-Douglas produtionfuntion:

Y _t = A _i,t K _i,t ^α L ^β _i,t

⁽⁴⁾

in logsand after addingthe error terms:

y _i,t = a _i,t + αk _i,t + βl _i,t + u _i,t

⁽⁵⁾

Variable

a _i,t

^{an be} interpreted as TFP,

u _i,t

^{are errors not related toTFP.}

It seems at rst that by estimating (5) using standard OLS, we an obtain TFP as

residuals from regression. Assuming that TFP is onstant through time, we ould also

estimatethismeasure usingxed eets panelregressions(suhalulationsforCentraland

Aording to Olleyand Pakes (1996) and later Levinsohn and Petrin (2003), estimating

rmlevelprodutivityusing OLSonaprodutionfuntionleadstoanendogeneityof fator

hoie problem. Omitting unobservable TFP in the estimation equation leads to omitted

variable bias - TFP is orrelated with fator hoie. Pavnik(2002) laimsthat using xed

eetspartiallysolvesthe problembutleadstoanestimatorofTFPthatisonstantintime.

Another partial solution is interating rm-spei dummy variables and a polynomialof

t

toaountfor TFP trends.

Olley and Pakes (1996) formulate a model, whih allows for onsistent estimators of

parameters of the prodution funtion and thus a onsistent estimator of TFP. It assumes

that the aumulationof apital isgiven by the followingequation:

K _t +1 = (1 − d ) K _t + I _t ,

⁽⁶⁾

where

d

^{is apital} depreiation. It means that investment at time

t

^{does not inuene}

apital in the same period. Olley and Pakes assume that produtivity observed by rms

a _t

has animpatoninvestmentinthesame period: the highertheprodutivity,the higherthe

investment. However, the funtional formof the relationshipis unknown:

i _t = i(a t , k _t ),

⁽⁷⁾

its inverse is of the form:

a _t = h ( i _t , k _t ) .

⁽⁸⁾

We an then write (5)inthe followingway (Arnold, 2005):

y _t = h(i t , k _t ) + αk _t + βl _t + u _t

⁽⁹⁾

y _t = βl _t + φ(i t , k _t ) + u _t

⁽¹⁰⁾

The above equation an be estimated by nonparametri methods or by a polynomial

approximationofthe unknown funtion

φ = αk _t + h(i t , k _t )

^{. Thisgivesaonsistentestimator}

of

β

^.

Firm makes its investment deision based on produtivity in time

t

^{and future prof-}

itability. Given that apital at time

t 1

^{is a funtion of investment in period}

t

^{, apital and}

produtivity are orrelated. Expetations onerning produtivity in the next period are a

funtion of produtivity in period

t

^:

E(a t+1 |a t , k _t ) = a _t+1 − ψ _t+1

^(where

ψ

^{is anerror). We}

an then write (Pavnik, 2002):

E(a t |a t− 1 , k _t− 1 ) = g(a t− 1 ) = g(h(i t− 1 , k _t− 1 )) = g(φ(i t− 1 , k _t− 1 ) − βk _t− 1 ),

⁽¹¹⁾

where

g

^{is anunknown funtionof}

φ

^and

k _t− 1

Substituting the aboveat

t

^{into(5)instead of}

a _t

^and reformulatingwe get:

y _t − βk _t− 1 = βk _t + E(a t |a t− 1 , k _t− 1 ) + ψ _t + u _t

⁽¹²⁾

= βk _t + g ( φ ( i _t− 1 , k _t− 1 ) − βk _t− 1 ) + ψ _t + u _t

The above equation an be estimated by non-linear method of

g

^{through a polynomial}

expansion of a funtion of

h

^and

k _t− 1

^{. Obtained}

β _k

^{together with}

β _l

^{an be then used to}

alulate TFP.

2.1 Data and estimation details

I estimate here a probit model of rms' export deision. The alulations were performed

on Polish rm-level data in manufaturing industry, olleted by Polish Central Statistial

for dierent thresholds of the share of exports in total rm revenue, to eliminate rms that

export onlyatinyshareof theirsales. Threedierentexportdeisiondummyvariableswere

reated: for rms whose exports were greater than zero and for rms whose exports exeed

1 and 2.5 perentof revenue.

The explanatory variables inthe model are the following:

•

produtivity (TFP[t-1℄)-thisvariableisestimatedusingthe OlleyandPakesmethod.

All data on apital, investment, employment and value added are taken from GUS

data. The proxy for apital is the value of xed assets. To aount for industry

tehnologyheterogeneity,TFPestimationswereperformedseparatelyforeahofthe2-

digitNACEsetors(greaterdisaggregationwasnotpossibleduetoinsuientnumber

ofobservationsinsomesetors. Theorretion forrms entryand exitwasperformed

using aprobit survival equation. Equation (12) takes the form:

y _t − βk _t− 1 = βk _t + g ( φ ( i _t− 1 , k _t− 1 ) − βk _t− 1 , P _t ) + ψ _t + u _t ,

⁽¹³⁾

, where

P _t = p(i t− 1 , k _t− 1 )

^the probability of survival until time

t

^{is a funtion of past}

investment and apital(see Pavnik 2002). This equation isestimated using NLS and

a thirddegree polynomial expansionof the unknown funtion

g

^.

•

exporter[t-1℄-laggedexportstatus. Thisvariablemeasurestheimportaneofthexed entryost ofexport partiipation. Ifthe obtainedestimatorispositiveandsigniant,

the presene of a rm in a export market is stable. Otherwise, the osts of entry are

signiantordoesnot havetobeinurredinsubsequententries ifthe initialentrywas

made (see Roberts and Tybout 1997).

•

^{rm size -this is measured by the logof} employment. Larger rms exploit eonomies ofsaletoalargerextendandanbemoreeetive. Moreover, given thesizeofoverall

osts of the large rms, the entry ost an berelatively less important.

•

^{foreignownership-adummyvariableindiatingmajorityofforeignownershipofarm.}

Foreignrms tendtofuntion assubsidiariesof multinationalsand their partiipation

in export markets reets the nature of their ativity as part of the multinational

struture.

•

^{state owned - adummy variable indiatingmajority of state ownership of a rm. On}

onehand,SOEareusuallyregardedasless eonomiallyeetive,beausethey tendto

havegoals other than pure protmaximization. Aording tothe theory above, these

enterprises shouldonaverage lessfrequently partiipateininternationaltrade. On the

other hand, in the ase of transforming eonomies, suh as Poland, SOE have been

present in the market longer that private rms and the osts of export partiipation

may have been inurred relatively earlier and do not play a signiant role (and the

ostsmayhavebeenalsoeasiertobearduetotheoldsystem'ssoftbudgetonstraint.

•

^{large - a dummy variable} orresponding to enterprises employing more than 500 em- ployees.

The followingsetoralvariables were alsoinluded.

•

^industry onentration - Herndahl index alulated using rm-levelrevenues data in eah 3-digit NACE industry. Firms operating in highly onentrated setors tend to

generate higher prots and it might be easier to them to bear the osts of export

partiipation. Moreover, havinglargemarketsharesinthedomesti marketmayallow

themtoross-subsidizetheirsalesintheforeignmarkettoseurebetterpositionthere.

Ontheotherhand,intensiveompetitionandlowonentrationmaypushrmstoseek

new opportunities abroad.

•

^importpenetration-aratioofimportstototalsalesinthedomestimarket,alulated using OECD (ITCS database) international trade data for 1996-2004 and sales data

from F-01 forms. An inrease in import penetration leads to shrinking prots and

pushesout rms intothe foreign marketorindues themtoexit the domestimarket.

•

^{tehnialbarrierstotrade(TBT)-adummyvariable. Sineall}traditionaltradepoliy instrumentsinnonagriulturaltradewerelargely removedintheproess ofintegration

withtheEU,whatisleftareinstitutionalbarrierstotrade. EUSingleMarketProgram

is targeting tehnial barriers totrade asmost importantsoure of remainingosts of

trade. Presen e of theEU poliyinapartiularsetor indiatesimportaneof TBT's.

Data on the EU poliy overage in the NACE 3-digit lassiation is taken from EC

(1998).

Unobserved timeand setoraleets are modeledthrough relevant dummy variables.

3 Results

3.1 Estimation results

Table 2 shows the results of prot estimations. These results have been obtained for rms

where exports exeed 1 perent of revenues. Estimations were made for all enterprises,

privateompaniesandonlydomesti ompanies. Resultsaremoreorless inlineforallthree

groups.

Pastexportstatusissigniantforallgroupsofrmsunderonsideration. Thisindiates

the existene of a mehanism desribed by Roberts and Tybout (1997). After entry to an

export market, rms presene is stable due tohigh entry and re-entry osts.

Table 3shows thealulatedmarginaleetsforaveragevaluesofvariables. Fordisrete

variables, the table shows eets of hange from 0 to 1. The results suggest that the prob-

ability of export in period

t

^{goes up by 77 perent if a rm was exporting at}

t − 1

^{. Past}

export statusis thus adominant fator driving the urrent export status.

TFP issigniantat 1perent level inallases underonsideration. This indiates that

the self-seletion to export market is present, whih is in line with theoretial literature.

This eetis strongerinthe groupof domesti enterprises thaninthe overall sample,whih

probablystemsfromtheweakersensitivityofexportstatusofforeignrmsduetothenature

Variable allrms private domesti

Exporter(t-1) 2.453 2.443 2.442

(107.45)*** (99.39)*** (101.33)***

TFP(t-1) 1.021 1.067 1.242

(3.91)*** (3.81)*** (4.30)***

Size 0.108 0.097 0.113

(log[employment℄) (4.23)*** (3.53)*** (4.11)***

Stateowned 0.087 0.077

(2.52)** (2.25)**

Large 0.079 0.100 0.046

(1.58) (1.80)* (0.85)

Foreign 0.478 0.486

(13.68)*** (13.81)***

Conentration 0.057 0.036 0.093

(2.22)** (1.27) (3.26)***

Importpenetration 0.184 0.188 0.133

(2.31)** (2.18)** (1.59)

TBT -0.218 -0.264 -0.154

(4.02)*** (4.44)*** (2.67)***

Constant -2.465 -3.044 -2.931

(11.13)*** (7.35)*** (12.28)***

Nofobservations 28365 24626 23449

Dummies:

Years YES YES YES

Setors YES YES YES

Estimationresults,

z

statistisinparentheses

*signiantat10%;**5%;***1%level

Variable Marginal eet X value

State owned 0,030 hange 0 -

>

¹

Large 0,027 hange 0 -

>

¹

Foreign 0,154 hange 0 -

>

¹

Exporter (T-1) 0,768 0,603

TFP (T-1) 0,362 1,009

Size 0,038 5,200

Conentration 0,020 0,507

Import penetration 0,065 0,286

TBT -0,068 hange 0 -

>

¹

year 1998 -0,077 hange 0 -

>

¹

year 1999 -0,093 hange 0 -

>

¹

year 2000 -0,034 hange 0 -

>

¹

year 2001 -0,028 hange 0 -

>

¹

year 2002 -0,048 hange 0 -

>

¹

year 2003 0,000 hange 0 -

>

¹

year 2004 0,071 hange 0 -

>

¹

oftheir ativity(dependentonexports andimports withinthemultinationalstruture). An

inrease of TFPby10 perent relativetoaverage ausesthe probabilityof export torise by

4 perent.

Size is signiant in explaining export status of rms. An inrease in the number of

employees from the average of 181 to 281 inreases the probability of export by 2 perent.

Variablelarge has nosigniantimpat onthe export deision.

Bothvariablesstateowned andforeign areimportantinexplainingthe urrentexport

status. As I mentioned before, state owned enterprises an have better position in foreign

markets due to their relatively longer history than private domesti ompanies. This may

also be a side eet of 1970s era of Gierek's industrializationwhere publi ompanies were

expanding rapidly enjoyingsoftbudget onstraintsand foreignloans abundantatthis time.

Foreign ompanies are involved in international exhange almost by denition. Marginal

eet ofstateownership is3perentand bythis fatorthe SOEshaveahigherthanaverage

probability of export. At the same time, the foreign rms export with probability greater

by 15 perentage pointsthan their domesti ompetitiors.

nies. The largerthe onentration, the higher isthe probability of exporting. However, the

marginal eet is rather low - a hange of the Herndahl index by 0,1 makes the export

deisiononly 0,2 perentmore likely. It ispossiblethat the size of the oeient isa result

ofexisteneoftwoompetingeets-pro-exporteetofmonopolisationandthepro-export

eet of ompetition. Import penetration is signiant, however, as in the ase of market

onentration, itseet on the probability of export is not very spetaular - aninrease in

penetration by 0,1 auses the probability of export to raise by 0,65 perentage points.

It seems that tehnial barriersto trade are important inexplaining the export deision

of rms. Presen e of any ofthe EU approahes totehnialbarrierstotrade (mutualreog-

nition, harmonization or new approah - essential requirements) dereases the probability

of exporting by 7 perent. This value seems rather large ompared to explanatory power

of other variables. However, it seems (or at least we ould hope for it) that it is not the

EU poliythat is atuallyausing barrierstotrade but insetors where thesemeasures are

present, the overall level of TBTis high. The expete d value of the oeient iseven lower

(higherin absolutevalue)if these measures were not in plae.

Marginal eets alulated for subsequent years shows a gradual inrease of the share

of exporters in the total number of rms. The probability of exporting between 1999 and

2003 inreases by 8 perent. Very important inrease of the number of exporters ourred

between 2003 and 2004. The probability inreases by another 7 perent in this time. This

an be ausedboth by the gradual dampening ofreession in2004 and the Polish aession

tothe EU that, ina step fashion failitatesentry toEU markets.

3.2 Produtivi ty and deision to export - sensitivity analysis

Subsequ ently, I analyze the sensitivity of estimates to the hoie of export threshold and

produtivity measure. Table 4 shows the estimation results with dierent export to total

revenue ratiothresholds (0perent, 1perent and 2.5 perent)and with alternativenotions

TFPw/seletion TFPw/seletion TFPw/seletion Laborprodutivity TFPw/oseletion TFPw/oseletion

Exportthreshold 1perent 0perent 2,5perent 1perent 1perent 1perent

Exporter(t-1) 2.453 2.107 2.530 2.451 2.452 2.454

(107.45)*** (91.24)*** (110.96)*** (107.31)*** (108.00)*** (108.08)***

TFP(t-1) 1.021 2.132 0.667 0.077 1.002 0.066

(3.91)*** (8.21)*** (2.59)*** (4.71)*** (3.89)*** (3.35)***

Size 0.108 0.130 0.117 0.173 0.178 0.118

(log[employment℄) (4.23)*** (5.21)*** (4.70)*** (8.97)*** (9.30)*** (4.64)***

Stateowned 0.087 0.123 0.051 0.085 0.089 0.089

(2.52)** (3.65)*** (1.48) (2.47)** (2.59)*** (2.60)***

large 0.079 0.123 0.057 0.076 0.081 0.083

(1.58) (2.13)** (1.17) (1.52) (1.64) (1.67)*

foreign 0.478 0.567 0.477 0.460 0.467 0.476

(13.68)*** (14.93)*** (14.19)*** (13.08)*** (13.51)** * (13.71)***

onentration 0.057 0.120 0.035 0.050 0.055 0.052

(2.22)** (4.08)*** (1.47) (1.96)* (2.14)** (2.04)**

importpenetration 0.184 0.159 0.174 0.214 0.186 0.186

(2.31)** (1.98)** (2.29)** (2.67)*** (2.34)** (2.34)**

TBT -0.218 -0.281 -0.177 -0.220 -0.201 -0.204

(4.02)*** (4.98)*** (3.40)*** (4.04)*** (3.68)*** (3.72)***

Constant -2.465 -3.545 -2.901 -2.119 -2.402 -1.820

(11.13)*** (16.20)*** (9.35)*** (11.75)*** (9.73)*** (10.98)***

Nofobservation 28365 28365 28365 28365 28640 28640

Dummyvariables:

Years YES YES YES YES YES YES

Setors YES YES YES YES YES YES

Estimationresults,

z

statistisinparentheses.

*Signiantat10%;**5%;***1%

of produtivity: laborprodutivity(ratio ofemploymenttovalueadded), TFP withoutor-

retionforrms'entryandexit,andabsoluteTFP(allpreviousalulationswereperformed

using TFP relativeto average ina given time periodand setor).

Results indiate some extent of sensitivity of the TFP variable oeient estimates to

the hoie of export threshold. When wetreats allrms had positive revenues fromexports

as exporters, the estimated oeient is almost twie aslarge as the one in the ase of a 1

perentthresholdandasthreetimesaslargeasintheaseofthe2,5perentthreshold. That

indiates higher level of produtivity among exporter rms than non-exporters, irrespetive

of the threshold. In all threeases estimated oeientis signiant and positive.

The signiane of state owned and large variables hanges with dierent export

thresholds. Stateownedenterprisesare,onaverage,haraterizedbyalowershareofexports

intotal revenues than private rms. On the other hand, largeenterprises havehigher share

of exports intotal revenues than remainingenterprises.

Use of labor produtivity instead of TFP as explanatory variable does not alter the

main onlusion so far. The estimate is signiant and positive. Obviously, the size of the

estimator is dierent than in the ase of relative TFP due to dierent onstrution and

variation of this variable. Similar onlusion may be drawn for the absolute TFP - it is

Explained variable Exporter

Export threshold 0 perent 1perent

H0: B[TFP(t-1)℄ = B[TFP(t-2)℄ = 0 4.98*** 5.51***

Explained variable TFP

Export threshold 0 perent 1perent

H0: Exporter(t-1) = Exporter(t-2) = 0 1.40 0.21

Ftest statistis,***rejet H0at 1 perent level

Thetable showstest statistisfor joint signianeof laggedexporter

and TFP variablesin explaining theirurrent values.

signiant and positive but annot be ompared to relativeTFP. We have to bear in mind

thatthe variationof absoluteTFP isdierentdependingonasetor(therewasa separate

produtionfuntionestimatedforeahsetor)andtheonlusionsdrawnmaystemfromthe

ross-setoral variation and not neessarily fromrm heterogeneity. Usingof the entry and

exit orretion inthe relative TFP estimation doesnot lead to large hanges inestimates.

The above results lead to a question: is rm behavior only a self-seletion into export

market, based on their urrent produtivity? Or maybe the model is inorretly speied

and the ausality isdierent: exporting leads to higherprodutivity.

Similarly as in (Arnold and Hussinger 2005), I seek for answers to that question using

the Granger ausality onept. I use a simple VAR model, where the explained variable

is produtivity (or export status) and on the right hand-side we have the lagged values of

produtivity and export dummy. The maximum lag is 2 periods, due to a rather small

numberof periodsinthe sample. The modelisestimated usingxed eetstoeliminatethe

risk of omitted variable bias.

Tests forsignianeof laggedexportstatus inexplainingtheurrentvaluesof TFPand

signianeof laggedTFP inexplainingthe urrentexport statuswere arriedout. Table 5

shows test statistisfor thenullhypothesis of noeet ofthese variablesonthe endogenous

variable. The results suggest that there exist a lear ausal diretion from TFP to export

deision(werejet

H 0

^{at1perentlevel). Atthe same timewe annotrejet thehypothesis}

of nolearning by exporting even at10 perent level.

✁

✂✁

✄✁

☎✁

✆✁

✝ ✁

✝✂✁

✝✄✁

✝☎✁

✝✆✁

✞

✄

✞✟ ✞

✂

✞

✝ ✠✡☛☞

✌

✝ ✂

✟

✄

✍✎✡

✏✑

✠✠✡☛☞

✌ ✒✓✑

☛✎

✔✑

✠✠✡☛☞

✌

Figureshows the deviation of produtivity ofrms entering exportmarkets

(inperent of standard deviation). Thiseet is purgedofyearand

setoral eets.

Figure3: Changes inexport afterentry

✕✖

✗✖

✘✖

✙✖

✚✖

✛✕✖

✛✗✖

✜✢

✣✤

✥

✛ ✗ ✦ ✘

Figure shows the share of exports in total revenues after entry to the

export market,purged ofyearand setoral eets.

thanks to interation with new markets, restruturing fored by foreign ompetition or by

knowledge spillovers abroad? I seek answers to that question by examining the paths of

produtivity of rms entering the export market.

Figure 2 shows hanges in produtivity of rms in the period of four years proeeding

export initiationand foursubsequentyears. Thisalulationswere separatelyperformedfor

rms who start exporting only one and for rms who start and stop exporting. Export

threshold was hosen at0 perent toeliminate rms whose export revenue osillatearound

a hosen threshold

Weansee,thatintheperiodsfollowingentry(intheaseofsingle-entryrms),theloal

maximum ofprodutivity(signiantlygreater thanthe average ofnon-exportingrms and

thaninthe period

t − 4

^{)oursatthe timeof entry. Inthe subsequentperiodsweobservea}

short drop in produtivity and in period

t + 4

^{we see aninrease in}produtivity that leads to a level higher than in any of the nine periods under onsideration. In the ase of rms

with multiple entries, the post-entrydrop in produtivity is lower.

The path in the export share of revenues (for single-entry rms) is shown on gure 3.

We an see, that sine the rst year of exporting, the share of exports inreases from 5 up

to 11perent infour years after export initiation. The average (amongall exporting rms)

export revenue share is 26 perent. Also, the produtivity of rms that are present in the

export market during all periods have a signiantly higher produtivity level than rms

that start exporting during the period under study. It seems reasonable to think that the

learningby exportingeetsaremoreoflongruntypeand starttoappearafterexports gain

a signiant share of total revenues. It may be the ase that identiation of these eets

with a 8-year sample is not possible.

This paper uses the Polish rm-leveldata to evaluate the determinants of export deisions.

The results obtained indiate an important role of produtivity in deision making. This

onlusion is irrespetive of the notion of produtivity used. What stems out from this

analysis is the existene of a self-seletion into export markets - more produtive rms

export with greaterprodutivitythan less eetiverms. Atthe same time,the importane

of lagged export status indetermining the urrent export status indiates existene of high

xedentryostintoexportmarkets. Itisalsoinlinewiththeintuition-tostartexportingit

neessarytoestablishontats inthedestinationountry,establisharetailnetwork,support

and servie enters et.

Estimationresultsalsoshowasurprising fatthatstateownedenterprises tendtoexport

more frequently than private rms. This may result from their, on average, longer history

and better experiene. Atthe same time, foreignrms export withgreaterprobability than

domesti rms.

Teststhatwerearriedout,seemtorejetthehypothesisoflearningbyexportinginfavor

ofthe selfseletionmehanism. Currentprodutivity isnotaetedbylaggedexport status

inaGrangersensebut urrentexport statusisindeedaetedby produtivity. Atthe same

time,thepathsofprodutivityofexportingrmsrevealasigniantinrease ofprodutivity

four years after entry into export markets. This may be an indiation of existene of two

paths ofausation: short term(fromprodutivity toexport) and longterm (fromexport to

produtivity). Formal veriation of this hypothesis needs longer samples and is learly a

eld for future investigation.

Referenes

Arnold, J. M. and Hussinger, K.: 2005, Export behavior and rm produtivity in german

manufaturing,Review of World Eonomis 141 (2),220239.

produtivity dierentials, turnover, and exports in taiwanese manufaturing,

NBER Working Papers 6235, National Bureau of Eonomi Researh, In.

http://ideas.repe.org/p/nbr/nberwo/6235.html.

Bernard, A. B., Eaton, J., Jensen, J. B. and Kortum, S.: 2003, Plants and produtivity in

international trade, Amerian Eonomi Review 93 (4),12681290.

Bernard, A. B. and Jensen, J. B.: 1997, Exeptional exporter performane: Cause, eet,

or both?, NBER Working Papers 6272, National Bureau of Eonomi Researh, In.

http://ideas.repe.org/p/nbr/nberwo/6272.html.

Bernard, A. B. and Jensen, J. B.: 1999, Exporting and produtivity, NBER

Working Papers 7135, National Bureau of Eonomi Researh, In.

http://ideas.repe.org/p/nbr/nberwo/7135.html.

Bernard, A. B., Jensen, J. B. and Shott, P. K.: 2003, Falling trade osts, heterogeneous

rms, and industry dynamis, NBER Working Papers 9639, National Bureau of Eo-

nomi Researh, In. http://ideas.repe.org/p/nbr/nberwo/9639.html.

EC: 1998, Tehnial barriers to trade, The Single Market Review, Dismantling of Barriers.

Sub-series III:Volume 1 ,OeforOialPubliationsofthe European Communities.

Luxembourg.

Estrin, S., Konings,J., Zolkiewski,Z. and Angelui,M.: 2002, The eet ofownership and

ompetitive pressure on rm performane in transitionountries. miro evidene from

bulgaria, romania and poland, LICOS Disussion Papers 10401, LICOS - Centre for

Transition Eonomis, K.U.Leuven. http://ideas.repe.org/p/li/liosd/10401.html.

Helpman, E. and Krugman, P. R.: 1985, Market struture and foreign trade : inreasing

returns,imperfet ompetition,and theinternational eonomy ,Cambridge,Mass.: MIT

Press.

Amerian Eonomi Review 70 (5),95059.

Levinsohn,J. andPetrin,A.: 2003,Estimatingprodutionfuntionsusinginputs toontrol

for unobservables, Review of Eonomi Studies70 (2),317341.

Melitz, M. J.: 2003, The impat of trade on intra-industry realloations and aggregate

industry produtivity, Eonometria 71 (6),16951725.

Olley, G. S. and Pakes, A.: 1996, The dynamis of produtivity in the teleommuniations

equipment industry, Eonometria 64 (6),126397.

Pavnik, N.: 2002, Trade liberalization, exit, and produtivity improvement:

Evidene from hilean plants, Review of Eonomi Studies 69 (1), 24576.

http://ideas.repe.org/a/bla/restud/v69y2002i1p245-76.html.

Roberts, M. J. and Tybout, J. R.: 1997, The deision to export inolombia: An empirial

model of entry with sunk osts, Amerian Eonomi Review 87 (4),54564.