Openness and Total Factor Productivity in Europe

(1)

M.Sc. International Economics and Business - University of Groningen

M.A. International Economy and Business - Corvinus University of Budapest

Double Degree Program

Master Thesis July 2013

Openness and Total Factor

Productivity in Europe

Laura Csorba S2418525

laura.csorba@gmail.com

(2)

2 Abstract

The productivity gap within the European Union is addressed by examining the total factor productivity (TFP) growth differences among member states during the period of 1995-2007. In the sample of 13 EU countries, including ones that joined in 2004, the effect of trade is estimated. A special openness measure is constructed to better capture the type of trade that truly contributes to TFP growth. It focuses on the intermediate input import structures, differentiating between imports from EU-members that joined before and after 2004. The results show that intermediate input imports from new member states have a significant positive effect on TFP growth.

(3)

3

Table of Content

I. INTRODUCTION ___________________________________________________________________________ 4 II. LITERATURE REVIEW ____________________________________________________________________ 6 III. HYPOTHESES ______________________________________________________________________________ 9 IV. DATA _____________________________________________________________________________________ 10

Dependent variable - TFP growth ______________________________________________________ 10 Explanatory variables - Openness measures ____________________________________________ 11 Control variables _____________________________________________________________________ 12

V. METHODOLOGY _________________________________________________________________________ 13 VI. RESULTS _________________________________________________________________________________ 15

Imports from new members ___________________________________________________________ 16 Imports from old members ____________________________________________________________ 17 Domestic inputs ______________________________________________________________________ 18 Imports and domestic inputs __________________________________________________________ 19 Limitations __________________________________________________________________________ 20

(4)

4

I. INTRODUCTION

The productivity growth slowdown of Europe has been a major concern of the Union. The productivity growth is not as fast as in the United States and the gap is widening further (van Ark et al. 2008). While trying to catch up with the world’s leading economy, less attention is given to the productivity gap within the EU. The countries that joined before 2004, referred to as the EU-15, are the ones considered when a comparison is made with the US. The Union now consists of 27 member states and this year a new member will join this heterogenous group of countries. These new member countries are lagging behind the EU-15, thus the internal productivity divide needs to be addressed, before the gap with the US is examined.

The convergence is an important goal for the union; substantial funds are spent on supporting the catch up of the later joined countries and underdeveloped regions. The identification of the main causes behind the increasing diversity is essential to deal with the issue effectively. After conducting traditional growth accounting, it is evident that the differences between countries are not due to differences in capital or labor stock. The root of the diversity is to be found in total factor productivity (Timmer et al. 2010). Easterly and Levine (2001) reached the same conclusion, saying TFP accounts for most of the growth differences over time, not factor accumulation. The output differences cannot be explained by the differences in available inputs (Ferreira and Trejos 2011). Total factor productivity (TFP) indicates the efficiency with which inputs are being used in the production process (Hulten 2001). It is a residual measure, meaning that it is the growth rate of output, not explained by the growth rate of inputs. TFP growth is a measure for disembodied technological change. It is influenced by organizational and institutional change. Since it is a residual, it also contains the errors and effects of unmeasured inputs (O’Mahony and Timmer 2009).

Several theoretical concepts are behind this residual. Hence a better understanding of total factor productivity is needed. As the Easterly and Levine (2001) paper on cross-country growth differences observes, factor accumulation on its own is not enough for achieving economic growth; it plays a role in igniting the development progress. Productive factor accumulation is what countries should aim for with policies and institutions focused on the determinants of TFP.

(5)

5 openness, competition, financial development, geographical predicaments and absorptive capacity (UNIDO 2007).

Here the context is Europe-specific and the focus is on countries of the European Union, from 1995 to 2007. This period covers the major enlargement in 2004 and the panel dataset includes countries that became members in this process. These years substantially changed the Union. From the several explanatory factors of TFP growth, a special attention is given to trade because of the countries under examination. The EU is integrated through trade more deeply than other regions in the world (World Bank 2012), which makes a special bond between these countries. Surprisingly, the role of trade in productivity convergence has not been addressed by studies in a European context before. The center of attention is usually foreign direct investment, when considering the determinants of TFP growth. The same attention should be given to trade, because trade and financial integration to the Union stimulated the convergence among the member countries, and the catching up to the more developed ones. Trade remains to play a fundamental role in sustaining economic growth in the Eastern European region (World Bank 2012).

Finding an appropriate measure for capturing openness within the EU is the most important part of this analysis. The openness measures used in the literature were often misleading. Simply taking export or import volumes compared to GDP does not capture the concept of openness. It matters what type of import products enter the country and in what form products leave. Trade volumes are not able to show how openness may contribute to TFP growth. The openness measures used in the literature are closely linked to income and productivity, which causes estimation problems.

To better examine the potential trade channels through which TFP growth can be affected, the production process should be considered. The availability of country input-output tables makes it possible to look at the intermediate input structures. Imported intermediate inputs are the type of trade which can truly contribute to TFP growth, thus a ratio is constructed to measure how much of the total intermediate consumption is imported. Going further, difference is made between where these inputs are imported from.

(6)

6

II. LITERATURE REVIEW

The existing literature, trying to find the explaining factors of total factor productivity growth, is very wide. A thorough review was done by UNIDO (2007), looking at micro, sectoral and macro studies and identifying the most important driving forces of TFP mentioned above. The UNIDO study differentiates between long term, deep determinants like institutions (political and economic), integration and geography; and medium term, proximate determinants such as capital formation and resource allocation. These cannot be strictly separated, because proximate forces work better in a supporting environment of deep determinants. Nevertheless, these long term drivers are wide concepts, their influence runs through several channels, having direct and indirect effects.

This review of the literature focuses on trade and related explaining factors of TFP growth. The expected positive effect of trade openness on total factor productivity growth has been proven in several studies. However, the significance and robustness of the results are not always convincing. The findings of the studies discussed here, give grounds to the hypotheses of this analysis on the European countries.

The empirical literature studying the effects of trade on total factor productivity has to deal with the reverse causality issue. It is possible that these two are simultaneously determined; meaning that openness increases productivity and the higher productivity induces trade flows, the causation runs both ways. For example, on the export side, productive industries are more competitive and are more likely to export. Positive changes in performance lead to growth in exports. This growth in exports induces more efficient allocation of resources, thus driving total factor productivity changes (Bernard and Jensen 1999). Trade liberalization may lead to increased competition, better access to intermediate products, export markets and technologies. Also, productivity enhancing effects of economies of scale and product specialization can be reached (World Bank 2007). The positive correlation does not mean causation; this endogeneity issue has to be dealt with, in order to get unbiased results on the effect of trade. The goal is to only measure the causality in one way and rule out the reverse effects, which may increase the estimated positive impact trade has on productivity. In the current analysis the openness measures are ratios, better capturing the trade structure and less affected by the endogeneity issue. Still, the relevant literature is reviewed to check how to deal with the endogenous trade variable and the usual problems encountered.

(7)

7 openness. He admits that measuring openness is a very complex issue, thus the positive relationship with TFP should be confirmed by examining the robustness of the results. The paper concludes that since the coefficients stayed positive throughout all the regressions, the effect of openness is confirmed. Trying to solve the reverse causality issue, he used historical lags as instruments. The problem with historical lags is that openness measures may be serially correlated, high openness measure affects next year’s value and the correlation runs through the whole time period. Thus to use lags as instruments is not a perfect solution, since they may not be exogenous. Still, Dollar and Kray (2003) used this approach, instrumenting openness variables via their lagged values and also in the current analysis this will be one way of dealing with the endogeneity issue.

The above mentioned problem with lagged values led to another line of studies, where instrumenting is done with geographical variables. As Frankel and Romer (1999) and Irwin and Tervio (2002) argued, geography is a powerful determinant of bilateral trade. The chosen geographical characteristics of a country affect income only through trade. Although these papers study the effect of trade on income, their estimation approach could be used in other cases, whenever trade is an explanatory variable in the model. Their main finding is that their instruments are truly exogenous. They estimate bilateral trade based on the country size, the distance from the trade partner, checking if they share a border and if one of it is landlocked. Then they use these predicted trade shares as explanatory variables in the regression to discover how trade affects income. A moderately significant but large positive effect of trade on income is found in Frankel and Romer (1999), which is confirmed in Irwin and Tervio (2002), who underline that countries that trade more have higher incomes. They argue that inclusion of country size to the estimation is important because larger countries may have better opportunities to trade within their borders, thus their trade share of GDP may be lower. The main drawback of using geographical instruments is that these cannot account for changes in the openness variable over time (Lee et al. 2004). It is an important aspect of the panel data in this analysis, thus this way of instrumenting will not be chosen.

Another issue in studies that deal with finding the explanatory variables of productivity is that trade ceases to be significant after controlling for geography, or if institutional quality proxies are in the equation (Irwin and Tervio (2002)). These results motivated the selection of control variables in this analysis.

(8)

8 The effect of openness on TFP growth differed in their sub-samples, for low-income countries openness only had a positive effect if human capital exceeded a threshold level. This finding shows that the countries’ development stage influences the results. Therefore, creating sub-samples in the panel data might lead to more accurate conclusions.

Alcala and Ciccone (2004) used ‘real openness’ as a trade measure (imports plus exports relative to purchasing power parity GDP). They criticize the total openness measure (imports and exports relative to GDP), which is unable to deal with the effect of cross-country differences in the relative prices of non-tradable goods. The argument is that productivity gains of trade are present in the tradable sector, leading to a relatively greater productivity, compared to the non-tradable sector. This will result in a rise of relative prices of the non-tradable goods and services. Therefore alternative measures are created as tradable GDP openness or real openness. Real openness means that imports and exports are included in the ratio in exchange rate US dollars, and the sum is compared to GDP in purchase power parity US dollars. The authors claim that better measurement of trade would show stronger positive effects on productivity. The instrumenting of trade flows is done as in Frankel and Romer (1999), and institutional quality is included following Hall and Jones (1999) focusing on government effectiveness and rule of law. In their study, geography and institutional quality are both controlled for, thus their regression equation suffers from multiple endogenous explanatory variables problem. They instrumented the endogenous variables and chose limited-information maximum-likelihood (LIML) estimator which is robust to weak instruments. In case the instruments in the current analysis are weak, the LIML estimator results will be presented.

Considering only Europe, Thorstensson (1999) analyzed the link between the European integration and growth through investment. He argues that integrated countries benefit from knowledge spillovers and finds that trade variables are important determinants of total factor productivity growth in Europe. The covered time period is from 1976 to 1990 and the included countries are 20 OECD members. It is expected that the benefits of the integration can be proven in a smaller time period and for a newly joined member countries as well. The paper considers initial TFP, investments (as a share of GDP), mean years of schooling, the growth in imports as a share of GDP and the growth in domestic R&D stock. A sensitivity analysis is conducted at the end, as the theoretical literature is not conclusive about which variables should be included in such regressions.

(9)

9 are not as heterogeneous, therefore the results might be Europe-specific. The key lesson learnt from the literature is that in order to achieve reliable results, controls for institutional quality and geography have to be introduced and the endogeneity issue has to be addressed when the effects of trade are analyzed.

III. HYPOTHESES

The UNIDO (2007) review summarizes that trade contributes to TFP growth through two channels. It has indirect effects through technology transfers and a direct influence on productivity growth, as it leads to more efficient allocation of resources across sectors (Ferreira and Trejos 2011). As regards the indirect effects, the key elements of technology transfer are knowledge creation, transmission and absorption, where trade works as a carrier of knowledge. Overall, the trade channel is found important but its significance depends on the absorptive capacity of the countries. In order to use the knowledge thus enhance productivity, a set of capabilities are needed. To account for these capability differences, empirical studies control for human capital and R&D investment when analyzing the effects of trade on TFP growth (UNIDO 2007). Some may benefit more from trade openness than others. Also, depending on the type of trade and its technological content, the effects can be diverse.

These findings lead to the expectation that within the European Union, trade openness could have an important positive effect on productivity. The effects could be more visible in the new member countries since the examined period includes the years after transition to a market-based economy, which entails important productivity enhancing changes. Moreover, the integration gives these countries easier access to more sophisticated intermediate inputs through trade liberalization. This is essential for the catching up process.

In the case of the EU-15 countries the benefits of the liberalization may occur through the optimalization of their imported input structure. The new member countries could be used as providers of cheaper intermediate inputs. The localization of production could change leading to greater economies of scale. The positive effect of openness will be tested on two levels.

Hypothesis 1: Openness had a positive effect on total factor productivity growth in the European Union in the period 1995-2007.

(10)

10

IV. DATA

Dependent variable - TFP growth

There are different approaches to derive total factor productivity growth. The three main methods and their assumptions are described in Jorgenson et al. (2007). In the current analysis the data are directly obtained from the EU KLEMS Growth and Productivity Accounts database (Timmer et al. 2008), which is organized around the growth accounting methodology and contains input data-series and derived measures. The database follows the production possibility frontier approach, which accurately summarizes the industry data. This approach is less restrictive than the aggregate production function approach, which builds on the existence of identical industry value-added functions. Still, it has its limitations since on the input side it assumes that each input receives the same price in all industries (Jorgenson et al. 2007). The derivation starts with the production possibility function, where the gross output is a function of capital (K), labor (L), intermediate inputs (X) and technology (T) for each industry j (time subscripts are suppressed for convenience)

, , ,

Assuming constant returns to scale and competitive markets, the value of output equals the value of all inputs. From a producers’ point of view the prices reflect the marginal cost paid by the user.

Going further, profit maximizing behavior is assumed. Following the derivation in the EU KLEMS database, TFP growth is defined as a translog index and equals the real growth of output minus the weighted growth of inputs.

∆ ln ∆ ln ̅ ∆ ln ̅ ∆ ln ̅ ∆ ln

The weights are the value shares of each input and averaged over two years

, ,

(11)

11 The database is based on national accounts statistics from which the data are processed and harmonized to ensure comparability. These refined estimates of total factor productivity1 are different from the TFP growth rates used by previous studies, which build up on own calculations or the OECD STAN database, usually following the aggregate production function approach. The EU KLEMS data consider labor, capital and intermediate inputs. In addition, the numbers are based on capital stock estimates better harmonized internationally (O’Mahony and Timmer 2009). For each country, TFP (value added based) growth data is a percentage change from the base year of 1995. The change in the TFP level is calculated each year by comparing it to the 1995 value. The comparability of these values is analogous to the consumer price index comparisons. The base year value might be different, but the pace of growth is under examination, from that specific year. The constraint of the panel dataset in this analysis is the availability of TFP growth data for EU countries in the database. The growth accounting exercise is not done for the whole union, especially the Eastern part of the new members, joined after 2004, is not covered completely. However, it is important to use comparable data, and the list of countries in the panel includes the most important economies of the EU-15. The EU-15 countries are referred to as ‘old’ members. One part of these countries is historically closely connected to Central Eastern Europe. The others are geographically distant and might not be as much affected by the Eastern enlargement of the union as the previous group. However, the trade openness measures used here could still be important determinants of TFP growth for all countries. The three new members examined are Czech Republic, Hungary and Slovenia. All of these countries went trough the transition process and dealt with the difficulties of reorienting their trade structure. During the years of opening up their markets to the European Union and the stabilization of the market economy institutions, all of these countries had the opportunity to radically improve their productivity numbers.

List of countries: Austria, Belgium, Denmark, Finland, France, Germany, Italy, Spain, The Netherlands, United Kingdom, Czech Republic, Hungary and Slovenia.

Explanatory variables - Openness measures

Various trade measures are used throughout the literature as discussed above. In this special case of focusing on Europe, a new measure is constructed to be able to better identify the transactions that influence total factor productivity. The importance of imports is that it is a channel through knowledge and technology is transferred (UNIDO 2007). More specifically, intermediate imports

1

(12)

12 define the production structures, thus contributing to productivity enhancement possibilities (Halpern et al. 2011).

The World Input Output Database contains data on the input structure of all European countries in a detailed industry level. After aggregating these into country level data, three aspects of the production structure are examined. These ratios are the main explanatory variables in the model.

Dom: Intermediate use of domestic output / total intermediate consumption

Old: Intermediate use of imports from old EU countries / total intermediate consumption New: Intermediate use of imports from new EU countries / total intermediate consumption Since these three ratios add up to almost 1 in this panel dataset, their effect on TFP growth will be estimated in separate equations. Imports from trading partners from the rest of the world are representing a small percentage in European countries (less than 10 percent in average), thus these are not built in the model.

Control variables

As the literature review of UNIDO (2007) emphasizes, controlling for geography is inevitable. Two variables account for this: (Area) – country size in km2 and (Pop) – the number of inhabitants2. The variable Area will drop out from the fixed effect estimation but will be a part of the 2SLS estimation method. Considering that these two geography related variables are correlated, population is replaced by a population density (Dens) variable in a later stage.

Additional controls are included based on the empirical findings that TFP growth is associated with the absorptive capacity of countries. Two major influential factors are considered here, human capital and R&D. The (Educ) variable is the hours worked by high skilled persons in the country3 and (RD) variable is the R&D expenditure of the country as a percentage of its GDP4.

2

Data collected for all countries and years from Eurostat database under Population and social conditions theme for area, population and density.

3

Share in total hours, data-series from the World Input Output Database Socio-Economic Accounts. The percentage for Total Industries will be used for every country and year.

4

(13)

13

V. METHODOLOGY

Arriving from the previous empirical work with panel data in this field, the main issues to deal with are finding an appropriate measure of trade openness, address the endogeneity problem and control for geography to achieve unbiased results. The models to be estimated are the following for all countries (i) and years (t):

∆ ! "#$ %&ln '(#) %*ln +, %-ln ./01 %2ln 34 5 (1)

∆ ! 67/ %&ln '(#) %*ln +, %-ln ./01 %2ln 34 5 (2)

∆ ! 4+8 %_&ln '(#) %_*ln +, %_-ln ./01 %₂ln 34 5 (3)

The first method of estimation is with fixed effects, which treats omitted variable bias by controlling for country-specific and time-constant influences (Hill et al. 2011). In the current analysis this method serves the estimation best, because the explanatory and control variables are not sufficient to fully explain the TFP growth, which has country specific determining factors. The omitted variables are controlled for by using the fixed effects method. The heterogeneity across individuals is taken into account, however the underlying assumption is that the differences between the countries are fixed. To be able to derive consequences from the results, the standard errors need to be robust. The results are not reliable if heteroskedasticity is not dealt with. Taking the logarithm form of the variables may solve the issue but a Breusch-Pagan test is conducted to check if the problem is still there. The null hypothesis, that the error terms have constant variances is rejected for the dataset with a p value of 0.000. Therefore, heteroskedasticity-robust standard errors are obtained for all estimations.

Alternatively, the estimation can be done with random effects, a technique which allows for random individual differences. In the same time, it is assumed that the error terms are uncorrelated over time and between countries and also uncorrelated with explanatory variables (Hill et al. 2011). These are very strong assumptions, which will probably not hold in the current panel dataset. A Hausman contrast test is carried out to check if the random effects estimator is inconsistent in this case. It tests the null hypothesis that the coefficient estimates are equal to another by comparing the estimates of the two different methods of estimation. The test is conducted for all three models and the null is rejected based on low p-values (see test results in Appendix). This means that the random effects estimator is inconsistent. Thus, the fixed effects estimator will give the reliable results.

(14)

14 term are not correlated (Hill et al. 2011). Based on the literature, it cannot be assumed in this dataset. To ensure the existence of endogeneity, a Durbin-Wu-Hausman test is conducted. The null hypothesis of no correlation between the independent variable and the error term is tested in the following way. The residuals are obtained from the OLS regression, where the potentially endogenous variable is regressed on all other explanatory variables and instruments. Then the fitted values of the residuals are put in the original regression equation as an explanatory variable and its coefficient is checked. In all three cases the coefficients of the fitted values are not significantly different from zero. Thus, the null hypothesis cannot be rejected and the existence of endogeneity is not proven.

Despite the fact that the fixed effects estimator would be sufficient for estimating these models, the unanimous suggestion of the literature is that the endogeneity issue is to be treated and previous studies had come up with several solutions. This analysis presents the results of the estimation via two stage least squares, the most common approach. To be able to conduct 2SLS estimation, valid and strong instruments are needed for the endogenous variable. In this case, the trade openness measures should be instrumented as they are potentially endogenous. A valid instrument is one that influences the endogenous explanatory variable and the same time is uncorrelated with the error term (Baum et al. 2002). If the model is exactly identified, one instrument replaces the endogenous variable. In that case, the latter condition cannot be tested. That is the reason why an overidentified model is needed, to be able to test the validity of the instruments used.

One way of instrumenting is to use the openness measure of a nearby country with similar economic size. Countries in the same geographic region may have similar trade structures and in this small sample it is possible to find a match for every country. Also, these measures are most probably uncorrelated with the error term in the original model.

To achieve instrumental overidentification, a surplus instrument is constructed with the openness measures’ respective lagged values. Before conducting the necessary tests, the data is checked for heteroskedasticity because its presence changes the reliability of our tests and estimation results. The Pagan-Hall test detects heteroskedasticity in the whole system, thus is more reliable than the previously used Breusch Pagan test in the case of instrumental variable regressions (Baum 2006). The p value of 0.000 indicates that the null hypothesis of homoskedaticity should be rejected (see Appendix).

(15)

15 instruments used because if they are only weakly correlated with the endogenous variable, the 2SLS estimator could be biased (Hill et al. 2011). At least one of the instruments is required to be strong. In order to decide on an appropriate estimator, the Angrist-Pischke multivariate F test is carried out. If the F-test statistic value is high enough (above 10 is the threshold) the null hypothesis can be rejected and the instruments are considered to be strong jointly.

When the equations suffered from the presence of weak instruments, the two stage least squares estimator gave biased results. The limited information maximum likelihood (LIML) estimator is more reliable in this case (Hill et al. 2011). However, this estimation method did not change the coefficients of the main explanatory variables, only the standard errors, thus the results are not presented.

Keeping in mind that the instrumental variable estimation is generally less efficient than OLS (Baum et al. 2002), the results from the estimation with fixed effects are preferred. That is the method that suits best the panel dataset in the current analysis.

VI. RESULTS

(16)

16

Table 1: TFP (value added based) growth 1995=100%

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Austria 99,8 99,4 101,3 102,8 104,8 103,7 103,5 103,7 105,1 106,2 108,9 111,2 Belgium 99,0 99,7 98,8 98,5 98,2 96,9 97,0 97,1 97,7 97,4 97,6 97,6 Czech Republic 99,9 95,5 92,4 92,1 92,8 94,6 94,6 96,1 97,6 101,1 105,4 107,8 Denmark 100,6 99,5 97,3 97,0 98,7 96,9 96,0 96,1 96,9 97,4 97,6 96,7 Finland 101,4 104,1 106,7 107,7 110,2 111,1 110,9 111,1 114,0 114,9 118,3 120,4 France 99,3 100,2 101,8 102,1 104,2 103,7 104,9 105,6 105,5 106,2 107,8 107,6 Germany 101,2 102,4 101,1 101,5 103,6 103,9 104,1 103,7 104,0 105,5 107,8 108,5 Hungary 105,8 111,9 115,0 115,0 118,2 122,5 124,6 125,7 128,8 131,1 134,6 133,7 Italy 98,9 99,6 98,4 98,0 99,4 99,3 98,0 96,5 96,7 96,2 96,4 96,6 Slovenia 101,5 104,3 104,7 104,0 103,6 106,8 107,9 106,1 107,5 108,7 110,4 110,3 Spain 98,1 97,6 96,6 96,7 96,2 95,6 94,8 94,1 93,4 92,4 91,9 92,1 The Netherlands 100,0 100,4 100,6 100,7 102,3 102,1 101,4 101,1 103,2 104,7 106,3 107,2 UK 101,1 101,2 101,2 101,2 101,5 101,5 101,3 101,1 102,3 103,2 104,3 105,1

The effects of imported intermediate inputs on total factor productivity growth were examined and the results of the three regression equations with different methods of estimation are presented below. All estimation results are for the whole dataset since the split up to sub-samples (new members and old members) did not lead to additional findings5.

Imports from new members

First, the effects of variable New, the percentage of intermediate inputs imported from countries that joined the EU after 2004, are examined. The table presents the coefficients estimated with fixed effects. To correct for the heteroskedasticity problem, regressions were run with heteroskedasticity robust standard errors, which in some cases turned the coefficients insignificant. In the case of the New variable, its positive effect stayed highly significant. One percent increase in the ratio of imported intermediate inputs from this set of countries would result in 5.17 percent increase in TFP growth. The R2 is close to 0.5 indicating the explanatory power of the equation. In Lee and Ricci (2001) the fixed effects estimation R2 values vary between 0.52 and 0.60.

After instrumenting, this positive effect and the model’s explanatory power decreased; also the Hansen J test rejects its null hypothesis that the instruments are valid. In addition, the country fixed effects are not controlled for; therefore the results of this regression with 2SLS could be misleading (test results are presented in the Appendix).

5

(17)

17 The control variables fail to produce significant results, however, the surprising negative coefficient of the population variable need to be addressed. It could be simply because of the dataset, where the countries with larger population had more modest TFP growth, or the population variable proxies for something different. This negative effect stays in all three equations, thus the variable was replaced by population density in the regression in the last two columns. This change reduced the explanatory power of the model and increased the positive effect of the New variable to 6.06, keeping it highly significant.

Table 2 Estimation results of Equation 1

(Fixed e.) (2SLS) (Fixed e.) (2SLS)

VARIABLES TFP TFP TFP TFP New 5.172*** 3.339*** 6.057*** 4.419*** (1.471) (0.636) (1.378) (0.628) lnpop/lndens -0.779 -0.0116** 0.00210 0.0268*** (0.545) (0.00532) (0.00348) (0.00814) lneduc 0.0673 0.0862*** -0.0156 0.125*** (0.0765) (0.0240) (0.0520) (0.0241) lnrd 0.0383 -0.0296 0.0233 -0.0646*** (0.0948) (0.0227) (0.100) (0.0246) lnarea 0.0151*** 0.00897** (0.00567) (0.00358) Constant 13.97 1.145*** 0.891*** 0.979*** (9.111) (0.0668) (0.110) (0.0595) Observations 169 156 169 156 R-squared 0.473 0.308 0.413 0.376 Number of countries 13 13 13 13

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Imports from old members

The variable Old is the ratio of intermediate inputs imported from EU countries that have been members of the union before 2004. These are mostly western European, developed countries, thus intermediate inputs from this region could be more sophisticated than the intermediate products from the new member countries.

(18)

18 member countries would increase TFP growth by 1 percent. The minor negative effect of density is also significant, on a 10 percent level; this could be explained by the probable bias in the sample, where the highest TFP growth numbers appeared in less dense countries.

VARIABLES TFP TFP TFP TFP Old 0.676 1.050*** 1.013** 1.247*** (0.389) (0.283) (0.442) (0.280) lnpop/lndens -0.732 -0.0213*** -0.00550* 0.0139* (0.551) (0.00633) (0.00298) (0.00835) lneduc 0.130* 0.0384* 0.0526 0.0548** (0.0665) (0.0217) (0.0469) (0.0214) lnrd 0.0854 -0.0157 0.0636 -0.0278 (0.0828) (0.0224) (0.0922) (0.0253) lnarea 0.0591*** 0.0508*** (0.0118) (0.0118) Constant 13.25 0.603*** 0.952*** 0.293* (9.192) (0.189) (0.127) (0.175) Observations 169 156 169 156 R-squared 0.379 0.276 0.343 0.262 Number of countries 13 13 13 13

The estimation with 2SLS worsened the R2 value, but the instrumenting technique worked better than in Equation 1. The Hansen J test of overidentification cannot reject the validity of the instruments (with a p value of 0.08 for the second column and 0.33 for the forth). The variable Old has a highly significant positive effect of 1.05 percent on TFP growth and 1.25 with the density control variable. Overall, this estimation seems to be less reliable than fixed effects, the interpretation of its results needs to be very cautious.

Domestic inputs

(19)

19 The R2 numbers improved compared to the other equations, meaning that this one explains TFP growth better. The coefficients stay highly significant in all cases. The negative effect of 0.88 % turns into 1.03 percent when the population control is changed to density. This replacement did not increase the fit of the model, but the expected negative effect stayed.

The coefficient lowered to -1.06 percent after instrumenting the Dom variable. In this equation the instrumenting technique improves the R2 value and the Hansen J statistic proves the instruments valid (with a 0.5 p value in the second column estimation and 0.62 in the fourth).

Overall this Equation 3 produced the desired test results and failed to reject the hypothesis from another perspective.

VARIABLES TFP TFP TFP TFP Dom -0.883*** -1.066*** -1.033*** -1.142*** (0.187) (0.131) (0.259) (0.126) lnpop/lndens -0.622 -0.0184*** -0.00342 0.0135** (0.491) (0.00483) (0.00392) (0.00628) lneduc 0.0592 0.0203 -0.0121 0.0348** (0.0623) (0.0186) (0.0471) (0.0166) lnrd 0.0249 -0.00577 0.00500 -0.0183 (0.0715) (0.0162) (0.0772) (0.0161) lnarea 0.0752*** 0.0649*** (0.00817) (0.00893) Constant 12.09 1.287*** 1.795*** 1.133*** (8.195) (0.0674) (0.174) (0.0620) Observations 169 156 169 156 R-squared 0.471 0.526 0.439 0.520 Number of countries 13 13 13 13

Imports and domestic inputs

In order to strengthen the conclusions drawn from the regressions, the variables New and Dom are considered jointly and estimated with fixed effects6. This model better explains TFP growth, as the improved R2 numbers indicate, and the results confirm the previous findings.

∆ !& "#$ !* 4+8 %&ln +, %*ln ./01 %-ln 34 5 (4)

6_{In the model where Old and Dom variables were considered jointly the variables lost their significance.}

(20)

20 Equation (4) with the population control variable is the best fit to the dataset, with 55 % explanatory power. The variable New stays highly significant, but the coefficient reduced by one percent compared to previous results. Based on these results, the increasing ratio of imported intermediate inputs from new member states improves TFP growth by 4.2 percent (or 4.5 when the density control is used). The ratio of domestic inputs in the production structure has a negative effect of 0.7 (or 0.8 with the density control) on TFP growth. These results of the positive effect of New and the negative of Dom are significant throughout all regressions in all models.

(Fixed e.) (Fixed e.)

VARIABLES TFP TFP New 4.154** 4.531*** (1.616) (1.471) Dom -0.706** -0.819** (0.274) (0.319) lnrd -0.0209 -0.0393 (0.0732) (0.0733) lneduc -0.00331 -0.0638 (0.0548) (0.0515) lnpop/lndens -0.505 0.00147 (0.419) (0.00375) Constant 9.888 1.484*** (7.049) (0.238) Observations 169 169 R-squared 0.550 0.528 Nr of countries 13 13

Limitations

The main limitation of the results is the small sample used in the analysis. This is due to data availability. The TFP growth numbers obtained from the EU KLEMS database were preferred over own calculations, because these are refined and reliable numbers. The preciseness of the data was more important than abundance. Augmenting the database to fully cover the new member countries of the EU would make it possible to examine Hypothesis 2. Unfortunately from the 13 countries in the sample, only three are new members.

(21)

21 countries’ trade relations, the enlargement of the EU. Still, it could be that the long term effects of this change are not captured in this short period of time. Also, these results cannot be contrasted to TFP growth before the sample years.

VII. CONCLUSION

Considering the results in light of the hypotheses drawn from the literature, it can be concluded that Hypothesis 1: Openness had a positive effect on total factor productivity growth in the European Union was supported by the results. The new measurement of openness used in the model eliminated the problem of endogeneity bias, a main issue in previous studies.

The positive effects of intermediate input imports from the newly joined countries seem stronger than imports from the old members, which means that TFP growth responds more positively to the openness towards the new member states in the EU. This is an interesting result considering the concern of increasing diversity within the EU. The more country joins, the more heterogeneous it becomes, but according to the findings of this analysis, having new members has benefits to the whole union. Connecting the new members into the production structures improves the TFP growth overall. This increase in TFP growth is responsible for better convergence among the countries. This result however may not be generalized; it only proves that the enlargement of 2004 benefited the EU as a whole in terms of TFP growth.

The hypothesis holds in another perspective as well. According to the analysis, having a low ratio of imported intermediate inputs results in lower TFP growth. The constructed variable tries to capture how closed countries are, examining the opposite side of the hypothesis.

(22)

22

REFERENCES

Alcalá, F. and Ciccone, A. (2004). Trade and Productivity. Quarterly Journal of Economics, Vol. 119, pp. 613-46.

Baum, C. F. (2006). An introduction to modern econometrics using Stata. Stata Press

Baum, C. F., Schaffer, M. E. and Stillman, S. (2002). Instrumental Variables and GMM: Estimation and Testing. Boston College Economics Working Paper 545.

Bernard, A.B., and Jensen, J.B. (1999). Exceptional Exporter Performance: Cause, Effect, or Both? Journal of International Economics, 47, 1–25.

Caselli, F. (2005). Accounting for cross-country income differences. In: Aghion, P., Durlauf, S. (Eds.), Handbook of Economic Growth, vol. 1. Elsevier. Chapter 9, pp. 679–741.

Dollar, D. and Kraay, A. (2002). Institutions, Trade, and Growth. Journal of Monetary Economics, Vol. 50, pp. 133-62.

Edwards, S. (1992). Trade Orientation, Distortions and Growth in Developing Countries. Journal of Development Economics, Vol. 39, pp. 31-57.

EU KLEMS Database (2008). Marcel Timmer, Mary O'Mahony & Bart van Ark. The EU KLEMS Growth and Productivity Accounts: An Overview. University of Groningen & University of Birmingham. downloadable at www.euklems.net

Eurostat (2013). Population statistics. Retrieved from:

http://epp.eurostat.ec.europa.eu/portal/page/portal/population/introduction

Frankel, D. H. and Romer, J. A. (1999). Does Trade Cause Growth? American Economic Review, Vol. 89, pp. 379-399.

Ferreira, P.C. and Trejos, A. (2011). Gains from trade and measured Total factor Productivity. Review of Economic Dynamics, Vol. 14, pp. 496-510.

Griffith, R., Redding, S. and Van Reenen, J. (2004). Mapping the Two Faces of R&D: Productivity Growth in a Panel of OECD Countries. Review of Economics and Statistics, Vol. 86, pp. 883-95.

Hall, R.E. and Jones, C.I. (1999). Why Do Some Countries Produce So Much More Output Per Worker Than Others? Quarterly Journal of Economics, Vol. 114(1), pp. 83-116

Halpern, L., Koren, M. and Szeidl, A. (2011). Imported Inputs and Productivity. CeFiG Working Papers 8, Center for Firms in the Global Economy, revised 16 Sep 2011.

Hill, R. C., Griffiths, W. E. and Lim, G. C. (2011). Principles of Econometrics. Fourth Edition. International Student Version. John Wiley & Sons

Hoover, K. D. and Perez, S. J. (2004). Truth and Robustness in Cross-country Growth Regressions. Oxford Bulletin of Economics and Statistics, 66 (5), pp. 765-798.

Hulten, C. R. (2001). Total Factor Productivity: A Short Biography. New Developments in Productivity Analysis, (vol. 63 of NBERStudies in Income and Wealth), ed. Charles R.Hulten, Edwin R. Dean and Michael J. Harper,1–47. Chicago: University of Chicago Press

(23)

23 Jorgenson, D.W., Ho, M.S., Samuels, J.D and Stiroh, K.J. (2007). Industry Origins of the American Productivity Resurgence. Economic Systems research, Vol. 19(3), pp. 229-252

Klenow, P.J., Rodriguez-Clare, A. (1997). The neoclassical revival in growth economics: Has it gone too far? In: Bernanke, Ben S., Rotemberg, Julio J. (Eds.), NBER Macroeconomics Annual 1997. The MIT Press, Cambridge, MA, pp. 73–103.

Lee, H.Y., Ricci, L.A. and Rigobon, R. (2004). Once again, is Openness Good for Growth? Journal of Development Economics, Vol. 75, pp. 451-72.

Miller, S.M. and Upadhyay, M.P. (2000). The Effects of Openness, Trade Orientation, and Human Capital on Total Factor Productivity. Journal of Development Economics, Vol. 63, pp. 399-423.

Miller, S.M. and M.P. Upadhyay (2002). Total Factor Productivity, Human Capital, and Outward Orientation: Differences by Stage of Development and Geographic Regions. mimeo, University of Nevada, Las Vegas

O'Mahony, M. and Timmer, M.P. (2009). Output, Input and Productivity Measures at the Industry Level: The EU KLEMS Database. Economic Journal, Royal Economic Society, vol. 119(538), pp. F374-F403, 06.

Prescott, E. (1998). Needed: A total factor productivity theory. International Economic Review, Vol. 39 (3), pp. 525–552.

The World Input Output Database (WIOD): Contents, Sources and Methods (2012)

Timmer, M.P. and Inklaar, R. and O'Mahony, M. and Ark, Bart van (2010). Economic Growth in Europe. Cambridge Books, Cambridge University Press.

UNIDO (2007). Determinants of total factor productivity: a literature review. United Nations Industrial Development Organization, Vienna

van Ark, B., O’Mahony, M. and Timmer, M.P. (2008). The Productivity Gap between Europe and the United States: Trends and Causes. Journal of Economic Perspectives, Vol. 22, pp. 25-44.

(24)

24

Appendix

Test results for Equation 1

Hausman test: Ho: difference in coefficients not systematic chi2(4) = (b-B)'[(V_b-V_B)^(-1)](b-B)= 26.16

Prob>chi2 = 0.0000

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance

Variables: fitted values of tfp chi2(1) = 64.61

Prob> chi2 = 0.0000

Pagan-Hall general test statistic : 54.684 Chi-sq(6) P-value = 0.0000 Ho: Disturbance is homoskedastic

Angrist-Pischke multivariate F test of excluded instruments: F( 2, 149) = 1358.54

Prob> F = 0.0000

Hansen J statistic (overidentification test of all instruments): 28.166 Chi-sq(1) P-val = 0.000

Hansen J for density equation: 27.914 Chi-sq(1) P-val = 0.000

Hausman test: Ho: difference in coefficients not systematic chi2(4) = (b-B)'[(V_b-V_B)^(-1)](b-B) = 12.46

Variables: fitted values of tfp

Prob> F = 0.0000

Hansen J for the density equation:0.965 Chi-sq(1) P-val = 0.3258

(25)

25 Prob>chi2 = 0.0414

Variables: fitted values of tfp chi2(1) = 21.33 density 11.32 Prob> chi2 = 0.0008

Prob> F = 0.0000