Master thesis

Master thesis

RUG Groningen

Economics and Business Faculty

International Economics & Business

“International R&D spillovers via migration and their

effect on country-level TFP”

Vasil Raykov (s2195224)

v.raykov@student.rug.nl

Thesis Supervisor: Dr. S. Krammer

Thesis Co-assessor: Dr. S.R. Gubbi

2

Abstract:

This thesis investigates empirically the impact of foreign technology spillovers,

embodied in migration flows, on country-level total factor productivity, through

a fixed effects panel model for a set of 25 advanced and transition countries

throughout the period of 1996-2009. Results show somewhat robust evidence

that migration is an important channel for the international dissipation of

knowledge. Conclusions on the importance of home-based R&D activity are

less robust for the chosen set of countries. Nevertheless both have been found to

have a positive effect on country-level productivity.

3

1. Introduction

In times of recession, more than ever, the topic of sustainable economic growth seems to gain focus. And as the new growth theory, as noted by Grossman and Helpman(1991) and Le (2012), among others, suggests, technological progress is the direct outcome of an innovation process involving cumulative research and development experience. Krammer (2010) points out that several studies have shown that the variation in economic growth rates among different countries is largely explained by “the difference in technological improvements, rather than human or capital accumulation.” (Krammer, Sorin “International R&D spillovers in emerging markets: The impact of trade and foreign direct investment”, The Journal of International Trade & Economic Development Vol. 19, No. 4, December 2010, p.591). The fact is however that very few rich countries have the resources, level of development and knowledge to innovate and create technology. The rest of the world, especially transition and developing countries, must rely mainly on the inflow of R&D from abroad. Keller (2010) calls these inflows “technological diffusion” and explains that this process occurs on two levels – through market transactions (patents, licenses, copyrights) and externalities. These externalities are known as technological spillovers.

By definition, as given by Grossman and Helpman (1992, p.16), technological spillovers mean that “(1) firms can acquire information created by others without paying for that information in a market transaction, and (2) the creators (or current owners) of that information have no effective recourse, under prevailing laws, if other firms utilize information so acquired.” R&D spillovers refer to the involuntary leakage, as well as voluntary exchange of useful technological information (Steurs, 1994). This can happen through transactions and links between firms, formal and informal contacts between CEOs, managers or employees, situated in different firms; employees changing their workplace, after acquiring certain technological information, bringing it their new employers, posted workers migrants or even students.

4

total factor productivity (from now on abbreviated TFP) accounts for much of the difference in incomes amongst countries world-wide. The reasons for these differences of the TFP levels of different countries can be traced back to what drives productivity. As later on will be explained, one of the potential determinants is foreign technology, which diffuses internationally via several channels.

In this master thesis I try to answer the question whether migration should be regarded as one important conduit for the international diffusion of technology. It also aims to contribute to the literature of international R&D spillovers by inspecting this spillover channel and what effect does the technology diffused through it have on country-level TFP.

The rest of this thesis is divided as following: Section two will provide a brief overview of the relevant literature; section three will describe the data used in this study; this is followed by explanation of the econometric model in section four; finally, in section five, concluding remarks are made along with considerations for future research.

2. Review of the relevant literature

2.1 Total Factor Productivity

“Total Factor Productivity (TFP) is the portion of output not explained by the amount of inputs used in production. As such, its level is determined by how efficiently and intensely the inputs are utilized in production. An important part of this is technological innovation, and as popular theory suggests, R&D development is what measures technological innovation.”

(Diego Comin, Total Factor Productivity, New York University, 2006 p.1).

Thus, TFP can be explained by variables different than just inputs. In the context of this thesis productivity growth is possible to be traced back to technological externalities (international R&D flows through migration). In other words, I expect countries that have the absorptive capacity to take advantage of foreign spillover inflows to experience a rise in their productivity level.

5

mentioned are human capital and trade openness, efficiency as well as home and foreign technology. Griffith et al. (2003) find empirical evidence of three sources for productivity growth: innovation, induced by R&D; transfer of technology; and absorptive capacity based on R&D.

Miller and Upadhyay (1997) and Danquah, et al. (2011) among others, point out in their findings that total factor productivity is strongly and positively affected by trade openness, as well as by the outward orientation of the country. One of the other often mentioned drivers of TFP in literature is human capital (Kneler and Stevens, 2006, Benhabib and Spiegel 1994, 2002 amongst others). Benhabib and Spiegel find robust evidence that human capital influences productivity growth, indirectly, through the rate with which laggard countries catch up to the leaders. However their proof of direct effect of human capital on TFP is not as robust. On the other hand Kneler and Stevens argue that technology is not simply “the part of output that cannot be explained”, but rather a result of a constant process of innovation and research. Their empirical results show that both human capital and R&D are significant determinants of productivity. They also confirm that absorptive capacity is indeed an important factor for efficiency. Even so, they argue that R&D spillovers do not explain the differences in TFP amongst countries, which is in contrast with most literature.

Nevertheless, in the light of this thesis, we will focus mostly on technology, both coming from domestic R&D efforts and through international inflows of knowledge, as a determinant of TFP

2.2 The link between R&D and TFP

As already noted before, the main channels for technology spillovers that have been studied so far in literature are trade; foreign direct investment (abbreviated FDI from now on);

distance from the technological frontier; and (to a lesser degree) international migration. In

this section a brief overview of the more relevant literature is provided.

A fundamental study in the field, without a doubt, is Coe and Helpman’s work

“Technological R&D spillovers”. Pooling panel data for 21 OECD countries plus Israel, Coe

6

was followed by various responses, building on it, as well as criticizing it. One notable work inspired by the CH paper is, written by Lichtenberg and Polaterie (from now on referred as LP). The authors aim to reexamine the work of CH, as they suggest an improved model that studies the link between imports and domestic TFP. On a separate note, LP study inward and outward FDI as potential channels for knowledge diffusion and find conclusive evidence that outward FDI, together with imports, influences productivity in a positive and significant manner. However, they also find that contrary on popular belief inward FDI is not a significant channel for technology transfer.

CH, together with Hoffmeister (CHH from now on) revisit their paper twice – in 1997 and again in 2008. They use an expanded dataset and modern estimation techniques, adding institutional variables to their previous model. Their results confirm their main conclusions about the importance of domestic and foreign R&D stocks on productivity from the first paper. Also they prove that a country with little or no innovative activity of its own, benefits from the R&D activity of its advanced trade partners. Additionally, they find that institutional differences in legal origin or patent protection, for example, are strong determinants of TFP and influence the strength of international R&D inflows. In a separate work from 1999, CHH study the North-South extent of technology spillovers. They examine a set of 77 developing countries, with little or no expenditure on R&D and how they benefit from the R&D efforts of their technologically advanced trade partners. They find conclusive results that developing countries enjoy larger productivity growth when they have access to a large foreign R&D stock. Additionally CHH find evidence that a higher degree of trade openness, especially with a bias towards advanced countries with large expenditure on R&D, as well as a higher education level of the workforce, bring in higher productivity gains. Another study based on the work of CH, looking at both trade and FDI as channels for international diffusion of knowledge is the one conducted by Hejazi and Safarian,1999 (from now on HS). Looking at spillovers from the G6 countries to the rest of the OECD HS confirm that trade and FDI are both positive and significant channels for the diffusion of technology. However they also find that once FDI is added to the model, the importance of trade as a spillover channel decreases, while the overall spillovers increase. In this way the authors prove their hypothesis that FDI is an important channel for bringing in spillovers and that spillovers through FDI are larger than those coming through the channel of trade.

7

studies on what effect knowledge, embodied in trade and FDI has got on productivity. Krammer inspects the way R&D spillovers affect emerging markets, comparing the results for transition and developed countries. His findings confirm that both trade and FDI are simultaneous and important determinants of productivity. Among his other results, he finds robust evidence that trade has a larger effect on TFP than FDI, as well as that human capital has a direct effect on productivity and helps with the absorption of new foreign knowledge. At the same time government expenditure proves to have a negative effect on TFP and the overall investment rate of a country has no effect at all. In line with theory are also Krammer’s conclusions that the richer Western countries gain more from domestic R&D activity, while the poorer Eastern countries gain more from foreign R&D stocks. Keller’s results also show that imports and FDI are significant channels for the diffusion of technology, while there is no proof that exports have the same effect. He also draws the conclusion that while technology is intangible and tacit in its character, and should be easy to “transport” world-wide, it is still local in nature and there is no one global pool of technology as of yet. Keller also notes that, while the importance of knowledge diffusion seems to be growing, a certain level of local investment is needed in order for a country to take full advantage of this. On a similar note, Madsen (2010) proves that developed countries achieve growth through their own R&D activities (innovation), while developing countries rely more on foreign R&D efforts (imitation). His results show that R&D intensity, as well as its interaction with technological distance, absorptive capacity based on education and the technology gap, all affect productivity in a positive manner. However, he points out that for a developing country to take advantage of the possible growth premium that is created from its distance to the technological frontier substantial investment in its domestic R&D is required. Griffith et al. (2003) also find empirical proof that confirms that the interaction between R&D and the distance from the technological frontier affect TFP in a positive manner.

2.3 Migration as a channel for technology transfer

8

economy. Both researchers find that such spillovers have a significant and positive effect on a country’s productivity.

In his paper from 2008, Le, argues with authors claiming that the international migration of high-skilled workers affects sending countries in an only negative way resulting in “brain drain”.1

Le argues that the temporary movement of people might lead to productivity gains in the long term, as migrants might acquire technological knowledge from work experience or additional training that they receive in their new country. Furthermore, he claims that this knowledge can be transferred in various ways to their home country even if they do not return. While his findings confirm once more that domestic R&D activity, as well as human capital have a positive effect on productivity, they do more than that. He finds conclusive evidence that the international migration of high-quality workers can be an important channel for technology transfer. This thesis both uses Le’s findings, to a certain degree, as a foundation but relaxes the assumption that only high-skilled migrants bring in positive technology externalities and looks at the migration base as a whole. A reason behind this new assumption is that one can argue that innovation might be found at all stages of the technological process and at all levels of the economic hierarchy, thus blue-collar workers are just as likely to bring in technology spillovers as white-collar workers, provided they have the sufficient level of knowledge.

2.4 Conclusion based on the literature

Ever since the new growth theory gained popularity, researchers have been looking to find what drives productivity. Numerous studies have found results that human capital, investment, openness to trade, efficiency and technology are among the main factors that affect a country’s performance. Authors that have looked at technology, as such a determinant of TFP have found evidence that the effects of both domestic R&D activity as well as from international diffusion are significant and positive.

Although researchers have largely focused on the effects of trade and FDI as conduits of technology transfer, and have, with few exceptions, perhaps, overlooked migration as such a channel, their studies have always shown positive and significant results for the link between R&D spillovers and TFP. This, as well as, the positive results, obtained by Park (2004) and

1

9

Le (2008, 2012) that migration of high quality labor force is a positive and significant channel for the diffusion of technology, leads us to the first two hypotheses of this thesis:

Hypothesis 1: International migration is a significant channel for diffusion of technology between countries.

Hypothesis 2: R&D spillovers that flow into home countries through the channel of migration affect TFP in a strictly positive way.

On the other hand, given previous results in literature and the fact that the set of countries of interest studied in this thesis consists of rather developed countries, as none of them can be attributed as “developing”, leads us to believe that home R&D efforts will also be a significant factor affecting productivity. This is the third hypothesis to be tested:

Hypothesis 3: Home-based R&D activity is a significant determinant of country-level TFP The final hypothesis to be tested in this thesis is interconnected with the others, as I set out to compare whether domestic or foreign technology has got a bigger impact on the inspected set of countries.

Hypothesis 4: Domestic R&D efforts are a larger and more significant driver of TFP than foreign technology spillovers

Now, let us turn our attention to the empirical part.

3. Data

3.1 Description of the dataset

10

with China and India are the three biggest spenders on R&D outside the OECD area. However, sufficient data on migration to the latter two countries is lacking and they were not included in this study. A list of the “receiveing” countries, as well as an overview of the variables used in the regression model can be found in table A1 in the appendix. A list of the “sending” can be found in table 3.

The time span of the study is 14 years, between 1996 and 2009. This period coincides with the efforts of some of the included countries from East and Central Europe to employ a market economy and pave their way to the European Union, something that also had an impact on their western counterparts, both in terms of economics and migration.

3.2 TFP

As noted earlier, TFP is a function of the growth rate of aggregate output, capital and labor, and the respective shares in production of the latter two. In this thesis, I calculate the TFP values following the approach suggested by Herzer (Herzer, 2010).

TFP is derived, as usual, from the Cobb-Douglas production function, which in its standard form is the following:

In order to calculate the TFP we use the Cobb-Douglas function in the following formula:

(2)

11

Output, here, is represented by GDP (in constant 2000 US dollars), while labor is measured in total workforce (the number of people in working age employed in the economy). Data on both is extracted directly from the World Bank development indicators (WDI) database. As for the capital stock data, here, I follow Herzer’s approach (Herzer, 2010) and construct the physical capital stock from real investment data (measured in gross capital formation in 2000 US dollars, available in the WDI database) with the following equation:

K

= K

(1 – δ) +

I

. (3)

Where Kt is the capital stock in the current year (t); ) Kt-1 is the capital stock of the previous

year (t-1), or the initial capital stock; δ is the depreciation rate of the capital stock (in this paper I follow Bosworth, 2008 and fix it at 6%); and It represents the gross fixed capital

formation in constant 2000 US dollars (also taken from the World Bank development indicator database).

In order to calculate the initial capital stock

K

Harberger, 1978 is used:

(4)

year of the period and δ is once more the depreciation rate of 6%. 3.3 Spillovers

As this thesis studies the effect that the transfer of knowledge embodied in emigration flows, has on home country i, spillovers are defined as the sum of the outcome of all migration flows from i to a set of host countries j on one hand, and the research intensity of these countries on the other. Research intensity itself is defined as the fraction of gross domestic expenditure on R&D to GDP. To put the definition of spillovers in another form:

∑

(5)

12

is the gross expenditure on research and development in a country j in year t; and

refers to the gross domestic product in country j at time t.

The main assumption behind this model is that, ceteris paribus, a country would enjoy a larger spillover inflow when a larger number of its citizens migrate to the set of countries given.2 Data on migration flows from the 25 home countries towards the 10 host countries is extracted from the OECD International Migration Database. Data on the R&D intensity is taken from the OECD Main Science and Technology Indicators (MSTI) database. Since the MSTI database offers ready-made data on research intensity no own calculations are made. However, I should point out that due to some missing data on home R&D for some of the “sending” countries in this study, there are missing observations, thus making the panel unbalanced.3

Also it is important to mention here that quality data on migration is rather difficult to obtain. Partly this can be explained by the fact that illegal immigration is still present and an issue for developed countries. Also, given the latest European legal rules on migration, it is somewhat difficult to trace all migration inside of the European Union. Another reason is that in retrospect there is still missing data on migrant movement.

3.4 Data on the other variables

The other variables that are used in the estimation process of this thesis are domestic R&D

effort, education, trade openness and unemployment. For domestic R&D efforts (whether it is

in millions or as a fraction of GDP), data is gathered from the Eurostat database. Eurostat works with both data from the OECD and data gathered from the national statistical agencies. Education data is derived from the Barro and Lee dataset (2010) as the average years of schooling for people aged 15 and above. As a second source, in robustness regressions, I use data on tertiary rate of enrollment as a percent of gross enrollment from the World

2 _{A possible miscalculation was found in the spillover flows in the spreadsheets coming applied to this thesis.} However this flaw does not change anything in the model. It does not affect the movement of values in the model and does not jeopardize the results and the conclusion in this paper. The mistake consists of the fact that GERD to GDP was used as a whole number instead of as a fraction, while migrants were used as a fraction in thousands, rather than as a whole number. This is fixed in the dataset by multiplying the migrant values by 1000 and dividing the intensity numbers by 100. This results in 10 times higher base values for spillovers, which are used for recalculating the logarithm values for spillovers. Once regressions were run with the new values we see that there was no difference in the results.

13

Development Indicators (WDI). Trade openness (at 2005 constant prices) is obtained from the Penn World Table v.7.1 (2012). Finally data on unemployment is extracted from the World Bank.

4. Econometric model

The main goal ahead of this thesis is to examine migration as a potential conduit for technological knowledge spillovers and the effect migration-based externalities can have on country-level TFP, comparing their effects on home-countries’ economies to those coming from domestic R&D efforts. In order to do that, this thesis uses an econometric framework similar to other work in the field (Coe and Helpman, Krammer, Le, Keller among others), but using spillovers through migration as a main regressor of interest in the model.

A panel data model is chosen, as is often the case in similar studies. An advantage of this type of model is that it can combine cross-sectional and time-series data and estimate the effects that migration-related spillovers have on TFP over time. Furthermore, it is expected that different countries will have different unobserved traits that may depend on various country-specific individual factors. In order to account for this, it is wise to use Fixed Effects estimations. To make sure that Fixed Effects is the best alternative for this model, I conduct a Hausman test of Fixed versus Random effects regressions. The result of this test shows that the model fails to meet the asymptotic assumptions of the Hausman test, rendering Random Effects estimations inappropriate for this case. Thus all regressions are run with country-specific fixed effects.

The basic assumption behind the model is that a country’s TFP is a function of both home and foreign R&D efforts. Or to put this in another form, the basic regression equation is:

(

) (6)

Where is the total factor productivity of country i in year t, is the domestic R&D

effort of country i at time t, and represents once more the spillovers that country i receives in year t, through the channel of migration.

14

in five year-averages, taken from the Barro-Lee database (Barro and Lee, 2010); trade

openness (Penn World Table v7.01, 2012) and unemployment (WDI, World Bank, 2012).

Literature has widely supported the former two variables, as important determinants of TFP. Human capital (in the form of education), is proven to affect the countries’ ability to absorbing foreign technology, while higher openness should increase the effect R&D spillovers have on productivity. The notion behind using unemployment in the model is that since TFP is largely dependent on human capital, the misuse, or inefficient use of the latter, may result in a higher unemployment rate and should affect productivity negatively. One might argue that while education controls for the quality of human capital, the unemployment rate could control for its quantity.

In its full state, the econometric model, used in this thesis, takes the following form:

(7)

Where is the logarithm value of TFP for country i in time t, represents the

constant term for country i in year t, denotes the logarithm value of the

migration-based spillovers for country i at t, is the logarithm of the years of schooling for i

in year t, measures the trade openness of country i at time t (as a ratio),

marks the percent of unemployment for i at t and finally is the error term.

5. Results and analysis

15

1.38 and 1.58. Similarly to the other variables of the model their values’ do not deviate far from the mean, implying an equal distribution between countries.

Table 1 Descriptive statistics

Variable Obs Mean Std. Dev. Min Max

TFP 350 6.137615 .5934739 4.729562 7.032815 Domestic R&D effort 311 6.91247 1.582653 2.783158 9.462739 R&D spillovers 350 5.06903 1.380344 .5488693 8.651802 Education 350 2.282086 .1189374 1.942912 2.54544 Trade openness 350 .9973942 .3371101 .3395179 1.934134 Unemployment 343 .0805714 .0409141 .021 .22

16

Table 2 Correlation between variables of main regression

TFP R&D spillovers Domestic R&D Education Trade openness Unemployment TFP 1.0000 R&D spillovers -0.2767 1.0000 Domestic R&D 0.6221 0.4240 1.0000 Education 0.0770 -0.0562 0.0878 1.0000 Trade openness -0.0869 -0.2715 -0.3598 0.4450 1.0000 Unemployment -0.4635 0.2281 -0.2204 -0.1147 -0.0626 1.0000 (Number of observations: 310)

To test for the presence of heteroskedasticity in the model, two tests are conducted. First a simple Breusch-Pagan test is conducted, followed by a White test for heteroskedasticity. The results of both tests show that the model is biased by significant heteroskedasticity. In order to deal with this problem all regressions are carried out with robust standard errors clustered by country. The model also proves to have a relatively high (adjusted) (at 0.69), which is a sign that the variables included explain the changes in TFP rather well.

17

Table 3. R&D data averages. Home countries (1996-2009)

Countries Average TFP values Average GERD Average R&D intensity (in %) Average R&D Spillovers Austria 766.17 4,585.702 2.166429 300.0121 Bulgaria 156.42 254.1763 0.495714 506.1799 Belgium 823.60 4,845.217 1.905 103.6954 Croatia 333.47 467.5156 0.88875 312.2781 Cyprus 461.28 46.98358 0.349167 15.78388 Czech Republic 292.11 1,959.165 1.232143 240.0699 Denmark 902.88 3,292.752 2.400714 109.1082 Estonia 255.46 144.0123 0.895 34.46403 Finland 801.45 4,079.997 3.304286 90.34033 Greece 553.93 1,124.318 0.5675 421.0581 Hungary 314.68 1,042.478 0.882857 506.9449 Ireland 963.31 1,332.521 1.254286 155.588 Latvia 201.45 101.1739 0.47 84.38694 Lithuania 275.52 229.7453 0.67 134.0117 Malta 493.95 33.277 0.48875 4.977638 Netherlands 222.10 7,910.511 1.902143 324.2126 Norway 1,091.61 2,465.463 1.622727 60.66913 Poland 323.60 2,339.56 0.611429 3,359.404 Portugal 449.29 1,404.835 0.872857 349.2699 Romania 148.10 630.5641 0.463571 1,677.296 Slovak Rebublic 325.14 366.5656 0.626429 291.24 Slovenia 485.58 489.0129 1.443571 51.43131 Spain 579.80 8,566.559 1.042857 309.8306 Sweden 905.77 8,623.925 3.651 143.6304 Switzerland 854.16 5,677.993 2.7675 155.2195 5.2 Econometric results

18

the 99% level. However, I will not dwell on the results of this regression, as it does not use the full list of variables.

19

Table 4 Results of regressions

(1) (2)

VARIABLES logtfp logtfp

lrd 0.242*** 0.111*** (0.0157) (0.0389) logspill 0.0651*** 0.0670*** (0.0139) (0.0219) logschool 0.234 (0.272) open 0.211** (0.0910) Unmpl -0.0130*** (0.00359) Constant 4.295*** 4.512*** (0.125) (0.699) Observations 311 310 R-squared 0.690 Number of countries 25 25

Standard errors in parentheses

*** significant at the 1% level, ** significant at the 5%, * significant at the 10%

5.3 Robustness

To check the robustness of the model, two of the variables are altered. Firstly, domestic R&D expenditure is replaced with domestic R&D intensity (expenditure as a ratio of GDP). By doing so I account for the size of the local economies, as well as for the adjusted this way value of innovation. The second variable that is introduced in the robustness regressions is

tertiary enrollment as a percent of gross education enrollment. This new variable, used as a

proxy for education in other works (Krammer, 2010), is introduced in the place of years of

schooling as a measure for human capital. I run three OLS regressions with FE – the first two

20

They remain significant at the 99% level and even slightly increase in coefficient in regressions (5) and (6). However the new proxy for domestic R&D activity seems to be insignificant in both regressions it is used in. Furthermore, with the introduction of the new education proxy variable, domestic expenditure also loses some of its significance and is only significant at the 90% level. Results from the three robustness regressions are reported in table A5 in the appendix. The outcome suggest that once we account for the size of the economy by including the relation between expenditure on R&D and GDP, domestic R&D effort loses its significance. One potential explanation for this we can find back in Table 3, reporting the average R&D data for the selected countries. As evident in the table, in about half of the countries, domestic expenditure on innovation accounts for less than 1 per cent of their GDP. Such low R&D intensity suggests that technology in these countries is most likely by imitation rather than innovation. This is supported by evidence available in the same table. Most of the aforementioned countries have rather large spillover flows. Without (good) data on what countries actually spend on imitation, however, this is difficult to prove without a doubt. Another possible explanation could be traced back to the missing observations. Better data might produce more robust results. As for the rather surprising significance level of tertiary education, one can make a case that this can be used as proof that the emigration of people, whether for work or education might result in a significant TFP growth bonus for a country. Another thing that may be deducted from this result is that a higher percentage of tertiary students improve human capital significantly, thus resulting in a positive and significant change in TFP.

6. Concluding remarks

21

technology. Furthermore, the effect R&D spillovers, obtained through the channel of migration, have on the home country productivity are always positive, if not large. These results are also in line with the findings of Le (2008, 2012) and Park (2004), who have been among the few researchers that have looked at international migration as a tool that brings in technology externalities. This study however, fails to find robust proof for the third hypothesis Even though it keeps its positive coefficient, domestic R&D effort loses significance in the robustness tests. This occurs once that we have accounted for the size of the countries’ economies. This change of results can be attributed to either missing data or the fact that, about half of the inspected countries report a low level of domestic R&D intensity, implying that they rely more on foreign flows of technology and imitation. The findings of this paper fail to give valid proof for the fourth hypothesis of the study as well. If anything, one might argue that they provide proof for the exact opposite assumption – at least for this set of countries, foreign technology flows prove to be more significant determinants of TFP than domestic R&D effort. Explanation for this once more can be traced to the low research intensity that most of the studied countries report. The fact that some of the control variables maintain a significance level in the model implies that technology does cannot solely explain the change in TFP. Some of this change must be attributed to other factors such trade openness or human capital.

Results of this study, however, should be approached with caution, as it clearly has its limitations. First of all, values for TFP are calculated under the assumptions of constant returns to scale and perfect competition, which is difficult to be found in reality. Furthermore, when calculating the TFP values total workforce is used as a proxy for labor capital, while a much better proxy suggested in literature would be hours worked (Herzer, 2010). However data on hours worked is limited. Second, data on R&D expenditure is not always perfectly available, which results in some missing observations in the dataset. Good data on migration is also rather difficult to obtain and not always available, which might lead to underestimation of the spillover flows. Another limitation on this thesis is the unavailability of data on spending on imitation. This limits our knowledge of what some countries actually do spend on technology. Also, it is worth pointing out that due to the presence of heteroskedasticity in the model, all regressions had to be carried out with country clustered robust standard errors, which also had its impact on the results of this study.

22

23

7. Bilbliography

Abramovitz M. (1986). Catching Up, Forging Ahead, and Falling Behind. The Journal of Economic History, 46 , pp 385-406 doi:10.1017/S0022050700046209

Aiyar, Shekhar et al. “A Contribution to the Empirics of Total Factor Productivity”, Draft Paper, August 12, 2002

Beine M. & Docquier F. & Rapoport H., 2001. "Brain drain and economic growth: theory and evidence," ULB Institutional Repository 2013/10449, ULB -- Universite Libre de Bruxelles.

Benhabib, Jess & Spiegel, Mark M., 1994. "The role of human capital in economic development evidence from aggregate cross-country data," Journal of Monetary Economics, Elsevier, vol. 34(2), pages 143-173, October.

Benhabib, Jess and Spiegel, Mark M., Human Capital and Technology Diffusion (December 2002). FRB of San Francisco Working Paper No. 2003-02

Bosworth, Barry, and Susan M. Collins. 2008. "Accounting for Growth: Comparing China and India." Journal of Economic Perspectives, 22(1): 45–66.

Carter Hill R., Griffiths W.E., Lim G.C. “Principles of Econometrics” 4th edition Coe D.T., Helpman E. “International R&D Spillovers” / European Economic Review (1995) 859-887

Coe, D., E. Helpman, and A. Hoffmaister. 1997. North–South R&D spillovers. Economic Journal 107, no. 440: 134–49

Coe, D., E. Helpman, and A. Hoffmaister. 1999. Are there international R&D spillovers among randomly matched trade partners? – A response to Keller. IMF Working Paper 99/18, Washington, DC: International Monetary Fund.

24

Comin D. “Total Factor Productivity” New York University and NBER, August 2006 Filipe J. and Adams G. (2005). "The Estimation of the Cobb Douglas Function". Eastern

Economic Journal 31 (3): 427–445

Danquah, Michael, Moral-Benito, Enrique and Ouattara, Bazoumana, TFP Growth and its Determinants: Nonparametrics and Model Averaging (April 5, 2011). Banco de Espana Working Paper No. 1104

Griffith, Rachel and Redding, Stephen and Van Reenen, John (2003) R&D and absorptive

capacity : theory and empirical evidence. Scandinavian journal of economics, 105 (1). pp.

99-118. ISSN 1467-9442

Grossman, G.M., Helpman E “Trade, knowledge spillovers, and growth”, European Economic Review, Volume 35, Issues 2-3, April 1991, Pages 517-526

Grossman, G.M. and E. Helpman (1992), Innovation and growth in the global economy, MIT Press, Cambridge (MA) / London (UK).

Harberger, A.C. (1978), “Perspectives on Capital and Technology in Less

Developed Countries,” in Contemporary Economic Analysis, ed. by Artis, M.J. and A.R. Nobay, London: Croom Helm.

Hejazi, W. and A.E. Safarian (1999), 'Trade, foreign direct investment, and R&D spillovers', Journal of International Business Studies, 30 (3), 491-511.

Herzer, Dierk, (2011), The Long-Run Relationship between Outward FDI and Total Factor Productivity: Evidence for Developing Countries, No 41, Proceedings of the German Development Economics Conference, Berlin 2011, Verein für Socialpolitik, Research Committee Development Economics

Keller, Wolfgang, “International Trade, Foreign Direct Investment and Technological Spillovers”, Handbooks in Economics, Volume 02, 2010

25

Krammer, Sorin “International R&D spillovers in emerging markets: The impact of trade and foreign direct investment”, The Journal of International Trade & Economic

Development Vol. 19, No. 4, December 2010, 591–623

Le, Thanh “R&D Spillovers Through Student flows, Institutions, and Economic Growth: What Can We Learn From African countries”, Scottish Journal of Political Economy, Vol. 59, No. 1, February 2012

Le, T. (2008). Brain drain or brain circulation: evidence from OECD’s international migration and R&D spillovers. Scottish Journal of Political Economy, 55, 5, pp. 618–36. Lichtenberg, F. et Van Potelsberghe De La Poterie B. [1998], “International R&D

spillovers: a comment”, European Economic Review, 42, p. 1483-1491

Madsen et al “Catching up to the technology frontier: the dichotomy between innovation and imitation”, Canadian Journal of Economics / Revue canadienne d’Economique, Vol. 43, No. 4

Miller, Stephen M., and Mukthi P. Upadhyay. "The Effects of Trade Orientation and Human Capital on Total Factor Productivity." University of Connecticut Department of Economics

Working Paper Series 07 (1997)

26

8. Appendices

Table A1. List of Host countries and their R&D effort. Comparison to OECD total

Country GERD in million 2005 constant $ Share in OECD GERD

Australia 11523.03 1.58% Canada 20132.69 2.75% France 38287.01 5.24% Germany 62892.41 8.60% Italy 17588.4 2.41% Japan 118690.7 16.24% Korea 27151.74 3.71% UK 32250.32 4.41% USA 308844.4 42.25% Total 9 637360.7 87.19% Total OECD 731031.4 100% Russia 16728.71 2.29%

Table A2. List of variables used in the regressions

Variable Definition Source

logtfp Residual form of the TFP, in log form Own calculations lspills Spillovers via migration in log form Own calculations

lrd Domestic R&D effort in log form Eurostat logschool Years of education in log form Barro & Lee

(2010)

open trade openness % Penn World Table

Unmpl Unemployment as % of total workforce World

Development Indicators loghomerd Domestic R&D intensity in log form Eurostat

alteduc Tertiary education enrolment as per cent of gross World Development

27

Table A3. Summary statistics of the variables used in robust regressions

Variable Obs Mean Std. Dev. Min Max

TFP 350 6.137615 .5934739 4.729562 7.032815 Spillovers 350 5.06903 1.380344 .5488693 8.651802 R&D intensity 311 .0212883 .6700462 -1.514128 1.418277 Tertiary Education 342 .5295184 .1808769 0 .9507 Trade openness 350 .9973942 .3371101 .3395179 1.934134 Unemployment 343 .0805714 .0409141 .021 .22

Table A4. Correlation between the variables used in robust regressions

logtfp Spillovers R&D

28

Table A5. Results of the robustness estimations

(1) (2) (3)

VARIABLES logtfp logtfp logtfp

logspill 0.0651*** 0.0704*** 0.0701*** (0.0225) (0.0190) (0.0183) loghomerd 0.0309 0.0160 (0.0442) (0.0450) logschool 0.496 (0.350) open 0.282*** 0.115 0.150* (0.0823) (0.0883) (0.0826) Unmpl -1.532*** -1.296*** -1.488*** (0.408) (0.318) (0.363) lrd 0.0790* (0.0432) alteduc 0.315*** 0.391*** (0.0940) (0.106) Constant 4.634*** 5.181*** 5.666*** (0.755) (0.276) (0.0911) Observations 310 306 306 R-squared 0.654 0.731 0.713 Number of BLcode 25 25 25