Content uploaded by Tomasz Brodzicki

Author content

All content in this area was uploaded by Tomasz Brodzicki on Dec 28, 2017

Content may be subject to copyright.

ARGUMENTA OECONOMICA

No 2 (39) 2017

PL ISSN 1233-5835

I. ARTICLES

∗

Tomasz Brodzicki*, Dorota Ciołek**, Katarzyna Śledziewska***

WHAT REALLY DETERMINES POLISH EXPORTS?

THE SEMI-MIXED EFFECTS GRAVITY

MODEL FOR POLAND

The goal of this article is to investigate the determinants of the intensity of Polish exports

with the use of the trade gravity approach in a panel data set covering the period 1999-2013.

In an attempt to obtain unbiased results we utilize the semi-mixed effect method using PPML

estimator as suggested by the recent empirical literature. In the basic version of the trade

gravity model, we take into account the standard factors suggested by the literature of the

subject. In its extended version, we control for additional factors including relative

endowments of factors of production, a technological gap as measured by TFP and relative

patenting performance, quality of institutions or impact of regional and bilateral trade

agreements or exchange rate volatility. In most of the cases, the coefficients for the traditional

gravity determinants are economically sensible and their impact on the dependent variable is

statistically significant. The gravity approach proves to be a robust tool for identifying the

determinants of trade intensity and thus for predicting future trade flows.

Keywords: trade gravity, Poland, panel data, semi-mixed effects, PPML

JEL Classifications: C23, F10, F14, F15

DOI: 10.15611/aoe.2017.2.01

1. INTRODUCTION

Gravity has become the most important tool in international trade

analysis over the years1. Other “flagship” trade theories are useful for

determining the ground of exports/imports performances (specialization

patterns and trade directions). However, it is gravity that enables to

**∗ Faculty of Economics, Economics of European Integration Division; Institute for

Development, University of Gdansk, Sopot.

*** Faculty of Management, Department of Econometrics, University of Gdansk.

*** Faculty of Economic Sciences, University of Warsaw.

1 The research was conducted and financed within the grant of the National Science Centre

(NCN) entitled “An analysis of international trade of Poland in the light of new trade theories.

Implications for economic policy in the crisis era” (2012/05/B/HS4/04209) chaired by

Krystyna Gawlikowska-Hueckel. Furthermore, we would like to thank Jakub Kwiatkowski

for his help with collecting the trade data.

6 T. BRODZICKI, D. CIOŁEK, K. ŚLEDZIEWSKA

determine and predict actual trade flows or economic consequences of the

establishments of preferential trade agreements. The gravity model is also

used in interpretations of non-trade flows between countries, i.e. migration

and FDI. The framework of the model is based on the analogy with the

Newtonian theory of gravity reflecting the relationship between the intensity

of trade between two partners, the size of their economies and the distance

between them. The initial lack of clear theoretical underpinnings was solved

in the 1980s (e.g. Bergstrand 1985). Recently the model went through a

remarkable evolution in terms of empirical specification (for instance the

heterogeneity theorem by Melitz (2003)).

After studying numbers of articles devoted to the issue of gravity, an

important conclusion is that authors are rather flexible in their selection of

variables (in particular independent variables) based on the context of their

particular analyses. Apart from standard variables such as distance and

economic potential, a set of specific variables is usually included. In

international trade, these can be for instance: economic integration

agreements, currency unions, access to the sea, historical circumstances as

specific links between former metropolises and their colonies, the common

language, common cultural heritage, etc. A very detailed description of the

possible variables utilized in gravity equations was presented by

Kepaptsoglou, Karlaftis and Tsamboulas (2010). A thorough examination of

the gravity theory, as well as empirical tools and methods, was given by

Head and Mayer (2014).

Gravity models are characterised by the possibility to interpret

coefficients as elasticities, which is a positive aspect – however, it may

somehow create interpretational problems of the models’ parameters. The

issue of the proper specification of the empirical model has been one of the

most important issues for the last twenty years (e.g. Matyas 1997; Egger and

Pfaffermayr 2003). Silva and Tenreyro (2006) in their seminal article raised

a problem that had been ignored by both the theoretical and applied studies.

In particular, they argued, that the logarithmic transformation of the original

model is not a relevant approach to estimate elasticities. The multiplicative

trade models with multiplicative error do not satisfy the assumption of the

homoscedasticity of the error term since there is a dependency between the

error term of the transformed log-linear model and the regressors, which

finally causes inconsistency of the ordinary least squares estimator or the

random and fixed effects estimator. As an alternative, the authors propose

the estimation of the gravity model in levels using the Poisson pseudo-

maximum likelihood estimator. Moreover, more recently Proença et al.

WHAT REALLY DETERMINES POLISH EXPORTS […] 7

(2015) suggested a semi-mixed effects method which relaxes the very strict

assumptions of the random effects model but keeps more restrictions than

the standard fixed effect model. Therefore we select the latter as the most

relevant approach to study the determinants of the intensity of bilateral trade

flows of Poland with its trade partners all over the world in in the period

1999–2013.

The rest of the paper is organised as follows. Section 2 reviews the issue

of the proper estimation procedure. Section 3 presents a model for empirical

analyses and describes the data sources utilized. Section 4 discusses the

estimation results. The last section concludes.

2. THE ISSUE OF THE PROPER ESTIMATION PROCEDURE

There has been plenty of theoretical and empirical studies focusing on

estimation procedures that fit the gravity models for bilateral trade using

cross-sections of panel data2. The discussion refers especially to the correct

econometric specification of the model as well as the appropriate estimation

method. The simplest form of the gravity equation for trade, relevant to

Tinbergen’s (1962) approach, states that the trade flows from country i to

country j, denoted by Tij, is proportional to the GDPs of the two countries’

(Yi, Yj) and inversely proportional to the distance between the countries, Dij.

However, it should be taken into account that the trade is not the same as the

physical force of gravity, at least because it depends on many uncertainties

connected with economic and social activities, therefore it should be treated

as a stochastic process.

Furthermore, Anderson and van Wincoop (2003) argued that the

traditional gravity equation is not correctly specified as it does not take into

account additional variables representing the specific effects of the exporter

and importer. That is why in most of the empirical studies the gravity model

is extended in order to include many other variables (economic,

socioeconomic, geographical) that could potentially impact trade flows. As

explained above, the set of variables consists of the binary time-invariant Dij

(similar to the distance), nonbinary time-invariant information Zij, and other

time-varying variables which may vary also over i, j or both (i, j), denoted as

Xijt. Additionally, within the panel data framework, it is possible to include

unobserved individual (i) and time (t) effects as well as effects for pairs of

2 A review of such analyses was presented for example by Proenca, Sperlich and Savasci

(2015).

8 T. BRODZICKI, D. CIOŁEK, K. ŚLEDZIEWSKA

countries (i,j). At the same time, all of them may be treated as fixed or

random. Such models are defined as three-way effect panel data models.

The stochastic version of the panel data gravity equation has the

following form:

12

0

ij t ij

Dv

ijt it jt ij ijt ijt

T Y Y ZXe

δη

ββγα

αε

++

=

, (1)

where vt are time effects which could account for business cycles,

η

ij are

unobserved heterogeneity effects, and

ε

it is the stochastic effort term.

Further,

α

0,

β

1,

β

2,

γ

,

α

,

δ

are unknown coefficients.

There has been a long tradition in the trade literature of log-linearizing

equation (1) and estimating the parameters of interest by OLS, using the

equation:

01 2

ln ln ln ln ln

ln ln

ijt it jt ij

ijt ij t ij ijt

T Y YZ

X Dv

αβ β γ

α δ ηε

=++ +

+ + ++ +

(2)

But, as Santos Silva and Tenreyro (2006) argue, the validity of the

procedure depends critically on the assumption that

ijt

ε

ln

are statistically

independent on the repressors. When the assumption is not valid, the OLS

estimation is inconsistent and biased. The problem has been ignored so far

both in the theoretical and in applied analyses. In reality, the issue of

heteroscedasticity is quantitatively and qualitatively important in most

gravity models even if fixed effects are controlled for. Hence the variance in

the error term

ijt

ε

in equation (2) depends on Yi, Yj, Dij and other variables,

this means that the expected value of

ijt

ε

ln

will also depend on the

repressors, violating the condition for consistency of OLS. This suggests that

this estimation method leads to inconsistent estimates of the elasticities of

interest.

Another problem connected with the estimation of the log-linearizing

gravity equation are zero trade flows – observed trade between some pairs of

countries is equal to zero because the countries did not trade in a given

period or the value of trade was very low. The existence of observations for

which the dependent variable is zero creates a problem for the use of the log-

linear form of the gravity equation. Several methods have been developed to

deal with this problem. The first simple approach is to drop the pairs with

zero trade from the data set, but then the results of the estimation are not

reliable as some information is omitted. The second method is to use Tijt+1

WHAT REALLY DETERMINES POLISH EXPORTS […] 9

as the dependent variable in the regression, however, rescaling of the data

leads to an inaccurate estimation of the results. The next solution, used for

instance by Frankel and Wei (1993), is to estimate the multiplicative

equation using nonlinear least squares (NLS), which is an asymptotically

valid estimator for equation (1). However, the NSL can be very inefficient in

this context as it ignores the heteroscedasticity in the data (see Santos Silva

and Tenreyro 2006).

As a solution to both of the estimation problems described above

(inconsistency of the OLS and zero flows), Santos Silva and Tenreyro

(2006) proposed the Poisson pseudo-maximum-likelihood (PPML)

estimator, which is often used for counting data. They notice that if

economic theory suggests that y and x are linked by a constant-elasticity

model of the form

exp( )

ii

yx

β

=

, the function

exp( )

i

x

β

can be interpreted as

the conditional expectation of yi given x, denoted

[ ]

E|

i

yx

. A possible way

of obtaining an efficient estimate of the parameters of interest is the use of

the pseudo-maximum-likelihood (PML) estimator based on some

assumption on the functional form of

[ ]

V|

i

yx

. Under the assumption that

the conditional variance is proportional to the conditional mean,

β

can be

estimated by solving the following set of first-order conditions:

1

exp( ) 0

n

i ii

i

y xx

β

=

−=

∑

. (3)

The authors claim that for this estimator the data do not have to be

Poisson and yi does not have to be an integer for the estimator based on the

Poisson likelihood function to be consistent. Additionally, the use of the

estimator eliminates the problem of zero flows3 Westerlund and

Wilhelmsson (2009) also emphasized that the PPML estimator is a solution

to overcome the problem of zero trade flow observations.

The estimating regression of gravity model using PPML method has the

following form:

01 2

exp[ln ln ln ln ln

]

ijt it jt ij ijt

ij t ij ijt

T Y YZ X

Dv

αβ β γ α

δ ηε

= + + ++

+ ++

(4)

3 It should be noted that there are some studies raising some doubt on the generality of the

estimator for empirical trade models. For example, Martinez-Zarzoso (2013) shows that in

several situations feasible GLS combined with the log-transformation can have a better

performance than PPML.

10 T. BRODZICKI, D. CIOŁEK, K. ŚLEDZIEWSKA

It should be noted that in regression (4) the time effects, vt, and country-

pair specific effects

η

ij are estimated as fixed effects which cause that some

of the time-invariant effects cannot be estimated.

Another extension of the gravity model estimation is proposed by Savasci

(2011) who suggests estimating equation (4) with the aid of a mixed effect

PPML, where the pair effects

η

ij are random effects to control for

unobserved cross-section heterogeneity. The obvious problem that occurs in

such a model is to prevent misspecification due to the independence

assumption for the random effects. But in the context of small area statistics,

Lombardia and Sperlich (2011) introduced a new class of semi-mixed effects

models. In terms of panel econometrics, one could say that they extended the

Mundlak device for random effects models.

In accordance with the results of the above discussion, the estimation

of all gravity models presented in the present article has been performed

using the PPML procedure with robust standard errors in Stata (StataCorp.

2011).

3. THE EMPIRICAL MODEL AND DATA SOURCES

In their classic article, Anderson and Wincoop (2004) used export shares

of trade partners in order to estimate the strength of gravity. The use of the

county-pair effect allowed to eliminate the potential bias of mutual

resistance described in the literature of the subject. An alternative approach,

however, can be utilized (e.g. Helpman, Melitz and Rubinstein 2008), in

which the values of total trade flows are utilized. In our study, the value of

exports from Poland to a given trade partner in million EUR is the explained

variable (

).

ijt

Export

The estimated empirical panel model with country-pair effects for total

export takes the following general form:

01 2

exp ln ln ln ln ln

ijt

jt ij ijt ij t ij ijt

Export

βY β D X Z v

α γ ρ ηε

=

+ + + + ++

(5)

where

j

Y

is the size of the partner,

ij

D

is the distance to partner, and

ijt

X

is

the conditioning set of variables describing bilateral trade relations.

The basic explanatory variables include the size of partner as measured

by the log of real GDP (real GDP) or log of the population (population) and

the log of the distance between trade partners (distance). The distance is

WHAT REALLY DETERMINES POLISH EXPORTS […] 11

proxied by geographical “as the crow flies” distance from Warsaw to trading

partner’s capital (in kilometres).

Two countries of similar size (as measured by real GDP) should trade

more than two countries of dissimilar sizes. Helpman and Krugman (1985)

have shown that the smaller the difference in the relative size of economies,

the larger the volume of mutual trade, and the greater intensity of intra-

industry trade. This is due to the fact that, as economies become more

similar in terms of their market size, the potential for overlapping demand

for differentiated products is enhanced.

In the present study, we adopt two different measures of similarity and

expect the obtained coefficients to be statistically significant and positive.

The first one, sim, is calculated using the following formula utilizing data on

GDP of Poland Yi and the trade partner Yj:

=1

(6)

Secondly, following Balassa and Bauwens (1988), we calculate the

difference in GDP of Poland Yi and its trading partner Yj, DIFEij, as follows:

= 1 +

(7)

where =

.

At the same time, two countries at a similar level of development should

trade more intensely that countries characterized by a significant gap in the

level of development. In accordance with the tradition in the empirical

literature of the subject, real GDP per capita can be treated as a rough

measure of the level of development. We adopt the following measure of

the gap in the level of development (rlf) and expect its coefficient

to be statistically significant and negative. An increase in the gap should

decrease the intensity of bilateral trade and thus negatively influence Polish

exports.

ln

ji

rlf ypc ypc= −

(8)

We furthermore utilize a number of dummy variables for adjacency,

border or preferential trade agreements. Utilizing data from PWT 8.0 we

further control for differences in factor endowments or differences in

productivity.

12 T. BRODZICKI, D. CIOŁEK, K. ŚLEDZIEWSKA

Data sources

We utilize Comext data set as a principal source of trade data. Comext is

a statistical database on intra-EU and extra-EU trade of goods managed by

Eurostat, the Statistical Office of the European Commission.

For the set of explanatory variables, we utilize a number of data sources,

the most important being the Penn World Tables 8.0 by Feenstra et al.

(2013). We also utilize data sets of World Development Indicators (WDI)

and Worldwide Governance Indicators dataset (WGI) compiled by

Kaufmann et al. (2010), both provided by the World Bank. The data for

patent applications come from the United States Patent and Trademark

Office (USPTO). The OECD migration database has been utilized in order to

account for the size of the Polish diaspora.

4. ECONOMETRIC RESULTS AND DISCUSSION

The estimation of the basic and extended specifications of the empirical

model has been performed using a semi-mixed effects method suggested in a

recent article by Proença et al. (2015) with a dummy variable for

membership in the EU (EU) serving as a clustering variable. The estimation

was carried out in STATA 12. The results are provided in Tables 1 and 2.

The analysis is carried out for 234 trade partners of Poland in the period

1999-2013. The explained variable is the value of exports from Poland in

EUR million. The usual zero adjustment is not necessary as we take into

account the level of exports and not the standard log of exports.

Various specifications of the model were tested. The number of

specifications shown in the article has been restricted for obvious reasons.

The results are not sensitive to the inclusion of time effects. As they do not

significantly increase the fit of the model we have decided not to present

them. The general fit of the model is high – explaining from 76 to 93 percent

of the variation in exports depending on the specification. The results are

robust to potential modifications.

In most of the analysed specifications the coefficients of traditional

variables such as real GDP of a trade partner, and the distance between

Poland and the partner are economically sensible and their impact on the

dependent variable is statistically significant. The intensity of Polish exports

decreases with distance to the trade partner, and increases with the partner's

size – larger countries tend to import more from Poland. As expected,

geographical proximity has been shown to be an important determinant of

WHAT REALLY DETERMINES POLISH EXPORTS […] 13

bilateral trade flows as it can be associated with lower transportation and

information costs.

The impact of membership in the European Union (EU) is clearly

positive and statistically significant. Poland exports more, ceteris paribus, to

partners from within the internal market of the European Union (free flow of

goods and services within the free trade area and common market rules). We

would like to stress here once again that EU is our clustering variable in the

semi-mixed effects setting.

The impact of the gap in the level of development as shown by rfl is

negative, as expected in most of the specifications. This is statistically

significant, however only in a few of the specifications of the empirical

model. All in all, Poland tends to export more to countries at a similar level

of development.

As the index of similarity (sim) is correlated with a log of real GDP, in

order to test the Helpman and Krugman hypothesis we include in model M2

a logarithm of the population (population) as a measure of partners size. The

coefficient of sim, however, is not statistically significant which is to some

extent surprising. The result, as was checked, to a large extent depends on

the utilized estimation method. This is also the case with DIME (M3).

Adjacency plays a significant role as expected. Poland exports more to

neighbouring countries – the impact of both border and border length is

positive and statistically significant at 1 percent level (M4 and M5). The

border effect has been thus positively verified.

The coefficient of the Eurozone dummy (euro) is positive but statistically

insignificant (M6). This could be due to the inclusion of a dummy for EU

partner countries. If we drop it, the coefficient of euro becomes positive and

statistically significant. The result yields support for the existence of the so-

called Rose effect (Frankel and Rose 2002). Taking into account that already

most of the Polish exports go to the Eurozone, as well as the gradual

expansion of the Eurozone, the costs of staying outside of the single

currency area in terms of the unutilized export potential could be judged as

rather high. We thus expect a significant increase in Polish exports to the

Eurozone after the adoption of the common currency.

Furthermore, Poland exports more (M7), ceteris paribus, to countries with

better quality of institutions as proxied by rule of law. We have selected this

proxy for the quality of institutions from the set of alternatives provided by

the Governance Indicators database of the World Bank as it was most

frequently utilized in the empirical analysis of development.

14 T. BRODZICKI, D. CIOŁEK, K. ŚLEDZIEWSKA

The impact of exchange rate volatility on Polish exports as measured by

the log of the standard deviation of daily exchange rates of PLN observed

over a period of a year (S volatility) is negative but not statistically

significant (M8). The impact of volatility to USD (xr), available from PWT

8.0 is however significant and has the presumed direction (M9). Less

volatility in the forex market clearly stimulates bilateral trade.

The size of the Polish migrant community as proxied by the log of Polish

migrants in a given partner country (diaspora) has a surprisingly negative

and significant impact on Polish exports (M10). The data is available only

for a limited number of partner economies from the OECD which can clearly

bias the results.

The next three specifications (M11–M13, please refer to Table 2) analyse

the impact on the intensity of Polish exports of the difference or the gap

between Poland and its trade partners in factor endowments, productivity

and technological sophistication.

First of all, the greater difference in the capital-labour ratio (dif K/L ratio)

has a robust and positive impact on the intensity of Polish exports. This

could point to the still important significance of traditional factor

endowments differences as postulated by the HO theory and its modern

extensions (Heckscher and Ohlin 1991) in explaining a significant fraction

of Polish trade relations.

Secondly, the greater difference in productivity levels as measured by

total factor productivity ratios (dif TFPratio) decreases Polish export

intensity. We could thus infer that Poland seems to export more to countries

at a similar level of productivity and thus technological sophistication.

Thirdly, the impact of the technological gap as measured by the log of

absolute difference in relative patent applications (patent applications per 1

million population) in the United States Patent and Trademark Office (dif

abs CUMP) is statistically insignificant. We treat the ability to the patent in

the USPTO as a rough proxy for technological sophistication and the

proximity to the world technology frontiers. We would like to stress that the

results are sensitive to the method of accounting for the gap and the method

of estimation.

Last but not least, the results concerning the impact of regional or

bilateral trade agreements need a longer comment (please refer to models

M14–M15). First of all, we have to take into account that most of the Polish

exports have an intra-EU nature with Eurozone’s member states (and in

particular Germany) playing the most important role. The trade flows within

the EU are regulated by common market rules. If we account for free trade

WHAT REALLY DETERMINES POLISH EXPORTS […] 15

areas (FTA), customs unions (CU) as well as economic integration

agreements (EIA) with extra-EU states, the impact of all is positive but

statistically significant only for FTAs. In interpreting the results we have to

remember that in our panel the first four years (1999–2003) is the period

directly preceding Poland’s entry into the EU and thus preceding the

adoption of common trade policy rules. At the same time, we have to bear in

mind that the FTA in manufacturing goods with the EU, the most important

trade partner, entered fully into force in January 2002.

If we extend the analysis to all Regional Trade Agreements (RTA) and

control for specific relations with Post-Soviet countries (post-Soviet), the

impact of RTA on the dependent variable is positive and significant at 5

percent level (model M15). The establishment of RTA has a robust and

positive impact on Polish exports.

In the last specification, M16, we account for the overall level of

competitiveness as indicated by the value of Global Competitiveness Index

calculated by WEF (GWCI) – the weighted index of twelve basic pillars of

competitiveness (Schwab 2014). The impact on the dependent variable of

interest to us is statistically significant and positive in accordance with our

expectations.

CONCLUSIONS

The goal of this article was to investigate the determinants of the intensity

of Polish exports to its trade partners (country level). The analysis was

carried out for 234 trade partners of Poland in the period 1999–2013 with the

use of panel gravity modelling. We utilized a newly suggested (Proença et

al. 2015) and superior semi-mixed effects panel approach with the Poisson

pseudo-maximum-likelihood estimator. EU membership (EU) played the

role of the clustering variable.

The gravity framework proved to be robust. The fit of the empirical

model was high. The impact of standard determinants of gravity was highly

statistically significant and in accordance with general expectations. The

hypothesis on the positive impact of size similarity was not supported.

Adjacency had a robust and positive impact on Polish exports thus

supporting the notion of the border effect. This applied as well to EU

membership and membership in the eurozone (responsible presently for most

of the Polish exports with the dominant role of Germany – Poland’s main

trade partner). The results gave support to the Rose effect.

16 T. BRODZICKI, D. CIOŁEK, K. ŚLEDZIEWSKA

In the extended version of the model, we controlled for additional factors

including relative endowments of factors of production, a technological gap

as measured by TFP and relative patenting performance in the USPTO (at

the global technology frontier), quality of institutions, exchange rate

volatility and the impact of regional and bilateral trade agreements.

Exchange rate volatility had, as had been expected, a negative impact on

exports. The greater difference in capital to labour (K/L) ratio had a robust

and positive impact on the intensity of Polish exports, which could point to

the significance of traditional factor endowments differences as postulated

by the HO theory. At the same time, Poland seemed to export more to

countries at a similar level of productivity and thus technological

sophistication.

Some of the obtained results were surprising, such as the negative impact

of Polish diaspora, and thus require further investigation. The present study

will be deepened and extended along several dimensions in order to account

better for the technological gap and sectoral heterogeneity.

REFERENCES

Anderson, J., Van Wincoop, E., Gravity with Gravitas: A Solution to the Border Puzzle,

“American Economic Review”, Vol. 93(1), pp. 170–192, 2003.

Anderson, J., The Gravity Model, “Annual Review of Economics”, Vol. 3(1), pp. 133–160, 2011.

Balassa, B., Bauwens, L., Changing Trade Patterns in Manufactured Goods: An Econometric

Investigation. North Holland, Amsterdam, 1988.

Bergstrand, J. H., The Gravity Equation in International Trade: Some Microeconomic Foundations

and Empirical Evidence, “The Review of Economics and Statistics”, pp. Vol. 67 (3),

pp. 474–481, 1985.

Egger, P., Pfaffermayr, M., The Proper Panel Econometric Specification of the Gravity

Equation: A Three-way Model with Bilateral Interaction Effects, “Empirical Economics”,

Vol. 28(3), pp. 571–580, 2003.

Feenstra, R. et al., The Next Generation of the Penn World Table, NBER Working Paper no.

19255, 2013.

Frankel, J., Wei, S., Trade Blocs and Currency Blocs, NBER Working Paper no. 4335, 1993.

Frankel, J., Rose, A., An Estimate of the Effect of Common Currencies on Trade and Income,

“Quarterly Journal of Economics”, pp. 437–466, 2002.

Gopinath, G., Helpman, E., Rogoff, K., The Handbook of International Economics, Vol. 4,

Elsevier, North–Holland, 2004.

Head, K., Mayer, T., Gravity Equations: Workhorse, Toolkit, and Cookbook [in:] Gopinath,

G., Helpman, E., Rogoff, K., The Handbook of International Economics, Vol. 4. Elsevier,

North–Holland, 2014.

WHAT REALLY DETERMINES POLISH EXPORTS […] 17

Heckscher, E. F., Ohlin, B. G., Heckscher–Ohlin Trade Theory, MIT Press, Cambridge, 1991.

Helpman E., Krugman, P., Market Structure and Foreign Trade, MIT Press, Cambridge, 1985.

Helpman, E., Melitz, M. J., Rubinstein, Y., Estimating Trade Flows: Trading Partners and Trading

Volumes, “Quarterly Journal of Economics”, Vol. 73, pp. 441–486, 2008.

Helpman, H., Imperfect Competition and International Trade. Evidence from Fourteen

Industrial Countries, “Journal of the Japanese and International Economies”, Vo1. 1,

pp. 62–81, 1987.

Kaufmann, D. et al., The Worldwide Governance Indicators: Methodology and Analytical

Issues, World Bank Policy Research Working Paper No. 5430, 2010.

Kepaptsoglou, K., Karlaftis, M. G., Tsamboulas, D., The Gravity Model Specification for

Modeling International Trade Flows and Free Trade Agreement Effects: A 10–Year

Review of Empirical Studies, “The Open Economics Journal”, Vol. 3, pp. 1–13, 2010.

Lombardía, M. J., Sperlich, S., A New Class of Semi-mixed Models and Its Application in

Small Area Estimation, “Computable Statistical Data Analysis”, Vol. 56, pp. 2903–2917,

2012.

Martinez–Zarzoso, I., The Log of Gravity Revisited, “Applied Economics”, Vol. 45, pp. 311–

–327, 2013.

Matyas, L., Proper Econometric Specification of the Gravity Model, “The World Economy”,

Vol. 20(3), pp. 363–368, 1997.

Melitz, M. J., The Impact of Trade on Intra

‐

industry Reallocations and Aggregate Industry

Productivity, “Econometrica”, Vol. 71(6), pp. 1695–1725, 2003.

Proença, I. et al., Semi–mixed Effects Gravity Models for Bilateral Trade, “Empirical

Economics”, Vol. 48(1), pp. 361–387, 2015.

Santos Silva, J., Tenreyro, S., The Log of Gravity, “The Review of Economics and Statistics”,

Vol. 88 (4), pp. 641–658, 2006.

Savaşcı, D., Three Studies on Semi–mixed Effects Models, Dissertation presented for the

degree of Doctor of Philosophy at the Faculty of Economic Sciences of the Georg–

August–Universität Göttingen, 2011.

Schwab, K., The Global Competitiveness Report 2014–2015: Full Data Edition. World Economic

Forum, Davos, 2014.

Tinbergen, J., The World Economy. Suggestions for an International Economic Policy.

Twentieth Century Fund, New York, 1962.

Westerlund, J., Wilhelmsson, F. (2009), Estimating the Gravity Model without Gravity Using

Panel Data, “Applied Economics”, Vol. 43, pp. 641–649, 2009.

Received: December 2015

18 T. BRODZICKI, D. CIOŁEK, K. ŚLEDZIEWSKA

APPENDIX

WHAT REALLY DETERMINES POLISH EXPORTS […] 19

Table 2

Results of estimation for Polish exports

(11)

(12)

(13)

(14)

(15)

(16)

distance

-1.655***

-1.641***

-1.642***

-1.594***

-1.619***

-1.663***

(0.0594)

(0.0659)

(0.0688)

(0.0717)

(0.0748)

(0.0729)

real GDP

1.009***

1.016***

0.992***

1.020***

0.992***

0.991***

(0.0301)

(0.0387)

(0.0360)

(0.0378)

(0.0375)

(0.0301)

rlf

0.0443

-0.0358

-0.0651**

-0.0231

-0.00935

-0.0147

(0.0287)

(0.0305)

(0.0318)

(0.0267)

(0.0313)

(0.0392)

EU

0.543***

0.756***

0.658***

0.797***

0.764***

0.604***

(0.0637)

(0.0770)

(0.0727)

(0.0830)

(0.0726)

(0.0780)

dif K/Lratio

1.682***

(0.147)

dif TFP ratio

-0.663***

(0.215)

dif abs CUMP

0.0274

(0.0168)

FTA

0.272***

(0.0767)

CU

0.0628

(0.121)

EIA

0.187

(0.114)

RTA

0.00912**

(0.00458)

Post Soviet

0.137

(0.0944)

GWCI

0.120**

(0.0554)

Constant

-2.806***

6.006***

6.051***

4.978***

5.447***

5.641***

(0.792)

(0.298)

(0.287)

(0.393)

(0.382)

(0.400)

No of obs.

2,125

2,125

2,125

2,125

2,125

736

R-squared

0.906

0.845

0.834

0.836

0.829

0.931

Note: All regressions carried out using semi-mixed effect ppml with EU as clustering

variable. * Significant at 10%; ** significant at 5%; *** significant at 1%. Dependent

variable – total exports in EUR million. A total number of observations (No of obs).

Source: own elaboration. Estimated using STATA 12.