Msc. International Economics & Business:
Master Thesis
Drain or Gain? Migration and Technology Spillovers in
OECD Countries: Counteracting the Localization Effect
of Technology Diffusion?
Groningen, August 2008
Author Supervisor Hanjo Lu Dr. Bart Los
Faculty of Economics Faculty of Economics University of Groningen University of Groningen The Netherlands The Netherlands
Abstract
In my master thesis I investigate if migration within the Organization of Economic Cooperation and Development (OECD) member countries to the G5 countries (France, Germany, Japan, UK, USA) strengthens positive spillover effects of research & development (R&D) investments from the host countries on the migrants home countries’ productivity by overcoming geographic and cultural distances. Using a sample for 19 OECD countries from 1995 to 2003 I try to apply the empirical model of Keller (2002) who found geographical distance to have a diminishing effect of technological spillovers on productivity. I was not able to reproduce his results neither did I find migrants to have a positive effect on domestic labor productivity, but rather some significant negative effects. Future studies should investigate the effect of ethnic Chinese or Indian migrant networks as these societies are catching up economically by technological change and are also more socially cohesive.
Keywords: migration networks, trade, foreign direct investment, technological change, R&D
TABLE OF CONTENT
SECTION 1: INTRODUCTION ... 4
SECTION 2: THEORY AND LITERATURE REVIEW ... 6
SECTION 3: RESEARCH QUESTIONS……… …….11
SECTION 4: HYPOTHESES... ……….13
SECTION 5: EMPIRICAL MODEL... 16
STATISTICAL METHOD ... 20
SECTION 6: SAMPLING AND DATA COLLECTION ... 24
SAMPLING ... 24
DATA COLLECTION ... 26
SECTION 7: DATA ANALYSIS ... 34
DIAGNOSTIC TESTS ... 34
ESTIMATION RESULTS AND DISCUSSION... 38
SECTION 8: CONCLUSION... 46
LIMITATIONS... 47
SECTION 1: INTRODUCTION
The opening of the labour market within the member countries of the Organization for Economic Cooperation and Development (OECD) has led to increased migration especially of the highly skilled. Students, researchers and professionals look for opportunities across the borders of their home countries, and their mobility is facilitated by lower movement barriers. In the OECD area as a whole, the share of people with tertiary education is higher for the foreign-born (23.6%) than for the native-born (19.1%) (OECD 2008). Despite marked differences across countries, this finding holds for most individual OECD countries. But what is the economic effect of migration on the source countries? Is there a “brain drain” as these people are missing in domestic production, especially, when they are highly skilled, or is there a gain as migrants might facilitate the access to knowledge resources in their host countries? Studies with regard to economic effects of migration have rather emphasized the loss of skilled labour and the negative consequences on economic growth in the home countries (Beine, Docquier, Rapoport 2001). In my thesis I will address the latter part of this question and investigate if a country might also benefit from some of their citizens emigrating.
countries, France, Germany, Japan, the UK and the USA has been identified to have influences on technical change and innovation in other countries. Several studies have given empirical evidence that domestic and foreign stock of knowledge by G5 countries both influence domestic productivity (Coe, Helpman 1995; Lichtenberg, Pottelsberghe de la Potterie 2001 et al.).
Knowledge being one of the main ingredients of innovation, is often tacit and is in most cases non-codifiable, cannot be formalized or written down. Hence, its transfer depends on the closeness between the actors. The localization of research activities is hereby an explaining factor of the degree of innovation and technological change in different countries (Audretsch, Feldman 2003).
Furthermore, differences in the culture of a region and relationships between actors may also contribute to differences in innovative performance across regions (Malecki 1997). International migration might minimize the effect of these differences. As knowledge is carried and spread by people in most cases through informal channels their increasing mobility could counter diminishing effects of distances from their home countries to technology intensive regions by direct interaction as well as by promoting mechanisms of product exchange.
discuss the theory and literature on innovation and economic growth and previous studies. From the gap in the existing literature I will derive my research question in section 3. In section 4, I will state my hypotheses. Section 5 shows the empirical model to test the hypotheses. Section 6 will deal with the data collection for the analysis. In section 7, I will discuss the results of the analysis and in section 8, I will draw my conclusion, stating the economic relevance, limitation of my study and give indications for future studies.
SECTION 2: THEORY AND LITERATURE REVIEW
training, in a process of trial and error and operating experience. This is in line with a study by Nelson (1990)1, who has shown that technology consists mainly of person (and institution) embodied tacit knowledge, which is mainly transferred through independent learning, personal contacts and mobility. Von Hippel (1994) stated that high context, uncertain knowledge (or “sticky” knowledge as he terms) is best transmitted via face-to-face interaction and through frequent contact. According to Manski (2000), many interactions in R&D which are important to endogenous growth theory occur on informal level and are influenced by expectations, preferences and constraints of economic agents. The implying difference between information and tacit knowledge is that the marginal cost of transmission of the first one has become location invariant due to technological revolution in telecommunication, while the marginal cost of transmission for the latter still is lowest by frequent social interaction, observation and communication (Audrech, Feldman 2003).
On macro level, the strength of social networks in a country is hence crucial for agglomeration of knowledge. According to Malecki (2001), one of the fundamental aspects of “high technology” is its reliance on people. As knowledge is geographically concentrated2 (Jaffe 1993), access to it depends on distance to the region, where it is
agglomerated (Malecki 1997).
Research in the field of international technology spillovers with regard to economic growth has been part of the “endogenous growth” and “growth and trade”
1 In a survey more than 600 industrial R&D directors in the US were asked about the importance of
different means of learning about competitors’ product innovations.
literatures. Authors such as Krugman (1984; 1987), Ethier (1982), Romer (1990), Grossman, Helpman (1991) and Young (1991) brought methods of modelling imperfect competition and technological innovation into international economics and endogenized technological innovation, allowing the rate and direction of invention to be affected by the global pattern of specialization and trade. I lack the space in this thesis to go into details of these theoretical models, but rather note their implications for empirical research. Most empirical studies on international R&D spillovers based on these models measured technological activity input in form of expenditures in R&D and used production functions. They calculate total factor productivity (TFP) residuals for a set of countries using data on capital and labour, which are regressed on aggregate domestic R&D investments and a weighted measure of external R&D investments, where the weights are measures of bilateral trade between the countries. The most widely cited and influential work has been done by Coe and Helpman so far (1995).3 Their analysis has
been among others redone for developing countries (Coe, Helpman, Hoffmaister 1997), the newly industrialized countries and trade in IT-products (Madden, Savage 2002), which find similar results. In the context of geography and of relevance for my research, Keller (2002) found in a study on 14 OECD countries that the effectiveness of the foreign R&D capital stock from the G5 on domestic productivity in non-G5 countries declines when the distance between them increases, but may be strengthened if the sender and recipient countries of technology share the same mother tongue. This study gives an
3 Using data from 21 OECD countries an Israel in the period 1971-1990 they find that foreign and domestic
indication that tacit knowledge is transmitted via individuals, which is crucial in the process of technological change as mentioned in the beginning of the section.
Due to the geographical concentration of technological knowledge firms lacking competitive advantage may also locate plants and branch offices in knowledge-intensive areas in developed countries to acquire new technologies and skills. Foreign direct investment (FDI) can hence serve as a channel for technology spillovers, when motivated by technological sourcing. The idea is that large inward and outward activities of multinationals and large R&D stock of the partner country have positive influence on spillovers from foreign R&D to domestic operations (Navaretti, Venables 2004). On aggregate level Pottelsberghe de la Potterie and Lichtenberg (2001) gave empirical evidence.4 The presence of FDI motivated by technology sourcing gives support to the localization of knowledge.
Ethnical social networks are shown to have a significant impact on bilateral trade (Rauch, Trindade 2002) and foreign direct investments (FDI) (Tong 2004), as these can serve as a bridge for domestic business people to the host countries. Migrants can provide market information to host countries and create trust, as there are economic agents at location which do not only speak the same language but share the same cultural values. Firms use middlemen located in the region’s entrepôts, which act as intermediaries between buyers and sellers. Hence, firms might use their networks to take advantage of
4 They analyze inward and outward FDI flows for 13 OECD countries between 1971 and 1990 finding a
technological sourcing. Diffusion through FDI is mediated by the supply of local skills and R&D capability (Mody 2002; Nabeshima 2002).
These studies however do not capture the effect of mobility of individuals, which can bridge the effect of knowledge localization. These networking effects on knowledge spillovers have rather been addressed when measuring intermediate technology output in form of patents and patent activities, which could also serve as economic indicators as they found a significant relationship to R&D expenditures (Griliches 1990).
they have not been able to account for the effect of knowledge spillovers on productivity on aggregate level.
SECTION 3: RESEARCH QUESTIONS
There is hence empirical evidence on the importance of proximity for the adoption of knowledge in terms of patent citations, but not accounting for the effects on productivity growth, as is the case for R&D expenditures. I found Keller’s study (2002) to be the only one which took social and geographical proximity into account when studying the effect of foreign R&D expenditure on a country’s productivity. Geographical distance alone however does not capture complex social relationships (Feldman 2002, Branstetter 2002). Besides spatial closeness also social, cultural and organizational proximities are important when trying to benefit from innovation (Malmberg, 1997).
Following from this, I suggest that a country might gain by its people migrating to technologically advanced countries as they facilitate access to knowledge in the host country, not only when directly involved in technological activities, but also by serving as a “social” bridge into that country. The general research question in my thesis is:
The first specific research question concerns the mechanism, of how a country might gain from its emigrants:
Do migrants influence international knowledge spillover effects on the migrants’ home country’s productivity?
If migration networks do influence the effect of knowledge spillovers on domestic productivity, is it that they overcome geographic distance? So the second specific research question is:
Do migrants overcome geographic distance in its effects on international knowledge spillovers on the migrants’ home country’s productivity?
Does national cultural difference influence the effect of technology spillovers between countries?
If national cultural difference does matter on spillover effects, it may be bridged by the presence of countrymen at location, where the knowledge is generated. Recent studies have dealt with the concept of social networks in the field of international economic and business studies (Aldrich, Zimmer 1987; Saxenian 1993; Hsu, Saxenian 2000; Rauch, Trindade 2002; Tong 2004 et al.). One of the common characteristics of such a social network is ethnicity, as its members are also easy to identify. Cooperation is based on trust and transaction costs are reduced in that way. The fourth specific research question is:
Do migrants bridge the cultural distance and strengthen the effect of technology spillovers between countries?
SECTION 4: HYPOTHESES
Previous studies have identified the G5 countries, France, Germany, Japan, the UK and the USA as the main producers of “economic” knowledge, measured in R&D capital stock from which spillovers arise as their expenditure in R&D is almost the tenfold relatively to the other OECD countries5. The empirical evidence of localization of R&D spillover effects has as mentioned in the literature review been given by Keller (2002). As he did his study for time periods from 1970 to 1995 and found out that the effects of
geographical distance decline over time, I state due to networking effects at play that the effect of geography still is relevant. I intend to do the analysis for time periods which follow the ones in Keller’s analysis and replicate his results.
Hypothesis 1: The more geographically distant a country from the (other) G5 countries
is, the weaker are the effects of G5 R&D capital stock on its productivity
Following the importance of social interaction on the transmission of tacit knowledge besides geographical also social proximity between economic agents are crucial. On national level the congruence of social values might hereby play an important role. Kogut and Singh (1988) found out that social cultural differences can affect choices of entering markets. Cross-border business transactions involve interaction with societal value systems which differ from the domestic one. The distance between different cultures might hinder the transfer of tacit knowledge when social interaction is limited by different values, making it difficult to establish interpersonal trust. Large differences in these values would make it more difficult for different cultures to communicate at the same level, even when speaking the same language. The second hypothesis is hence:
Hypothesis 2: The more socio-culturally distant a country from the (other) G5 countries
is, the weaker are the effects of G5 R&D capital stock on its productivity
International mobility among OECD is more and more focused on the highly skilled, due to the internationalization of tertiary education and the importance of gaining international experience for professionals. International mobility has become an intrinsic part of the relationship between OECD countries. Mobility is facilitated by the presence of free mobility areas, agreements for recognition of foreign qualifications and easier access to visas (OECD, 2008). The migrants might form social networks in the host country with economic consequences. Besides direct exchange of the tacit components of R&D migrant networks can also trigger the spillover of knowledge by facilitating trade and direct investments from their source countries.
These social networks can help to overcome information barriers and promote trade and FDI by creating trust, providing market information and reducing transaction costs, promoting trade, investment and technology adoption in the country of origin (Docquier, Lodigiani 2006). Migrants might hence generate beneficial feedback effects for their countries of origin.
possible channels for R&D spillovers, which can be positively influenced by the sheer presence of countrymen in the host country. The third hypothesis is:
Hypothesis 3: The more migrants live in a G5 country, the higher will be the technology
spill over effects on domestic productivity in the country of origin.
The fourth hypothesis will depend on the outcome of the second hypothesis. If social cultural distance has a deflating effect on R&D spillovers, people immigrating to the G5 countries could bridge the cultural gap.
Hypothesis 4: The more migrants live in a G5 country, the weaker will be the minimizing
effect socio cultural distance on technology spillovers on the migrant’s home country’s productivity.
An important limitation is that the sheer number of migrants might not fully reflect the “tightness” of networks. More migrants from the same home country do not necessarily lead to more network or stronger networks, as I assume.
SECTION 5: EMPIRICAL MODEL
find that increasing distance between countries minimizes the effect of R&D spillovers on a country’s productivity. I chose to base my model on his as this serves my objective to investigate how cultural distance and migrants would affect productivity on aggregate level, as this is the intention of my thesis. Other alternative approaches to estimate knowledge spillovers use some measure of innovation other than output growth on the left hand side of the equation, for example counts of patents, of which the drawbacks I already stated at the end of section 2. This approach has been pursued by Hall (1993), Jaffe and Traijtenberg (1993; 1996).6 Another approach has been to apply a cost function
rather than a production function which has been pursued by Bernstein (1994), which considers the impact of R&D not only on total R&D but also on the amount of labour and intermediate products used. The reasons for me not to choose this approach is that it is performed on firm-level, and in line with Branstetter (2000) requires use of good input data price data which varies across units of observation and over time, which do not exist for R&D and physical capital even at the industry level. I continue with the model by Keller (2002).
Based on that the total manufacturing sector is positively related to domestic and foreign R&D, the effectiveness of foreign R&D is negatively related to the distance from the foreign economy. He uses an exponential functional form:
ln Lct = c + t + lnSct + ln( g G5 Sgt e- Dcg)]+ ct (1)
6 According to Branstetter (2000) Jaffe and Traijtenberg have been working on using direct measures of
As can be seen the model is nonlinear in its variables as well as in its parameters. The log-log function as applied in many studies serves explaining percentage changes in the dependent variables by percentage changes in the explaining variables.
This model will test the first hypothesis. L represents labour productivity, c is the index for the country, t is the index for time. S denotes accumulated investments7 in R&D, g is an index for the group of G5 countries, and Dcg is the bilateral geographic distance between country c—the technology recipient country (which in my thesis can also be a G5 country) — and the G5 countries, the technology senders. c,t, , , and are the parameters to be estimated, and ct is an error term. c captures the differences between countries, t the differences in time periods, e- Dcg captures the distance term, where is
the distance parameter, is related to the elasticity of productivity with respect to domestic R&D, determines the strength of the foreign R&D effect on productivity. The coefficient in the exponent indicates its influence on the explanatory effect on . If foreign R&D in the G5 raises productivity ( positive), positive estimates for suggest that these effects decrease with distance.
As socio-cultural distance might have the same minimizing effect on technological spillovers referring to the second hypothesis the model for testing this effect is:
ln Lct = c + t + ln Sct + ln( g G5 Sgt e- CDcg)]+ ct (2)
The term (CDcg) captures the socio-cultural distance between the two countries. is the
social cultural distance parameter. For positive values of the effects of R&D from the G5 on Lct decreases.
To test hypothesis 3 I will use the number of immigrants in the G5 countries as a new explaining variable in the regression
ln Lct = c+ t+ ln Sct + ln( g G5MIGcgt)+ ct (3)
The term ( g G5MIGcg) captures the number of immigrants of the technology receiver country in technology sender country. is the parameter indicating the possible strength of migrant networks. For positive estimates of the effect of technological spillovers to the source country increases. The term of including Sf is replaced by the term including migrants, as the degree of labour productivity will vary by the number of migrants in the G5 countries. Hereby, the assumption in the expression is that a higher number of migrants increase positive spillover effects on labour productivity.
For the fourth hypothesis I add cultural distance as variable, moderating the effectiveness on the number of migrants on domestic productivity Lct.
For positive values of the effect of immigrant networks will increase on the diffusion of technology.
To test these hypotheses nonlinear least squares regression will be used to measure the strength of the parameters within a small dataset. The nonlinearity of the model might lead to problems when carrying out the analysis. This is because the estimation results of nonlinear models depend on the starting values as in contrast to linear models iterative estimation procedures are used. Bad starting values will lead to an immediate halt of the computation without yielding any results as no convergence can be reached for the estimator.
Statistical method
cross-sectional as they enable to control for omitted variables that vary across states but not over time or the other way around. Furthermore more observations give more information.
The panel analyses models can be distinguished in static and dynamic models. Static hereby means that all relationships are between variables at the same point in time. I will explain and discuss the different models. I note that I indicated my choice of using a static fixed effects model already in the equations (1) to (4) by the application of the two , one for time t and one for countries c. For a dynamic model I would have to include time-lagged effects, due to the short time period I will however only use a static model. I first describe the fixed effect model than the alternatives to indicate the reasons why I did not decide to choose for them. The following section is based on Yaffee (2003) and Brüderl (2005).
Fixed effects model (Least square dummy variable model):
This model has constant slopes but intercepts that differ at the cross sectional units e.g. the country as in this analysis. There are no significant temporal effects but significant differences between countries. An important aspect of the model is the use of i-l dummy variables, where 1 is used to designate the particular country. That is why the model is also labelled “least square dummy variable model”.
There is a variety of fixed effects models. One could have constant slopes and intercepts that differ according to time. The model would have no significant country differences but might contain the risk of autocorrelation (which will be explained later on p. 38/39) owing to time-lagged temporal effects. In another model the slope coefficients are constant, but the intercepts vary over country as well as over time. Further models might have intercept and slopes either varying to the country or even varying to time and country as well. This sort of model would not only require i-l dummy variables for countries but also i-t dummy for time as well. The main benefit of the fixed effects model is the reliance on within estimation (all between estimation is lost) eliminating time-constant unobserved heterogeneity. However, with the fixed effects model we cannot test the effects of time-constant variables. Hence, a serious drawback of this model is that all explanatory variables that do not vary within the cross-section unit (time-constants) are lost, because they are picked up by the intercepts.8 Another drawback of the model is that
too many variables would lead to a decrease in degrees of freedom leading to a loss in the power of statistical test. Further a disadvantage is, if there are too many cross section units of observations they would require too many dummy variables for their specification and have a higher risk of multicollinearity and autocorrelation. This is however not the case in my model. The cross sectional fixed effects are of interest in my thesis, as my goal is to analyze differences across countries rather than across time. Furthermore, I only have data over a short time period, so using a static fixed effects model will allow me to control for volatility in cross sections and periods with both cross section and time effect dummies.
8 For this reason the coefficients for the distances cannot be estimated in a linear way when employing
Constant coefficients model:
This is the simplest one of the models. The constant coefficients hereby refer to intercepts and slopes. If there are no significant cross sectional or temporal effects, data could be pooled and ordinary least squares models could be run. Although most of the time there are either country or temporal effects, there are occasions when neither of these is statistically significant. The key assumption hereby is that there are no unique attributes of individuals within the measurement set, and no universal effects across time.
Random effects model
This model assumes that the intercept is a random outcome variable. The random outcome is a function of a mean value plus a random error
Yit = 1i + 2X2it + 3X3it + eit (6)
1i = i + i
SECTION 6: SAMPLING AND DATA COLLECTION Sampling
The variables can be categorized in country specific and inter country variables. The country specific variables are labour productivity and R&D expenditure; the inter country relationship specific ones are geographical and national cultural distance and the number of migrants. The number of migrants is relationship specific, because I am interested in their origin as well as their destination in order to estimate the effect from their source country to their home country. The dependent variable in this analysis is labour productivity in the OECD countries; the explanatory variables are domestic and “effective” R&D stock from the G5 countries. Explanatory variables on the “effectiveness” of the G5 R&D stock are geographical distance, social cultural distance and the number of migrants in the G5 countries. Keller (2002) used total factor productivity (TFP) to test how R&D spending levels in the G5 countries relates to the productivity levels in other 9 OECD countries, when influenced by geographical distance. In my analysis I extend the total sample to 19 countries, the G5 and 14 other OECD countries. Further in my analysis even G5 countries can be classified as technology receivers by other G5 countries.
especially relevant for cross country comparisons as it is the case in this study. Labour productivity based on value added is also used as a common measure for productivity.
The sample will be collected for the period from 1995 to 2003 for the 19 OECD countries Australia, Belgium, Canada, Czech Republic, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Korea, Netherlands, Norway, Poland, Spain, Sweden, UK and USA, representing locations on four different continents, with two countries from the former communist block (not counting East Germany). As these two countries joined the OECD in the 90s together with Korea, which joined in 1995, there is no data industrial data before 1993 and 1995 respectively for these countries. This also explains the short length of the time series, as earlier data is not available for some countries. The advantage of an OECD sample lies in their comparability, as they all have a certain standard of economic and political development, and intra OECD migration is not motivated by political reasons (e.g. asylum seeking). Furthermore a large share of intra-OECD migration consists of highly skilled which is crucial in the process of technology spillovers as mentioned.
Data collection Productivity
For my purpose of calculating labour productivity I use an internationally comparable dataset on industrial performance at ISIC industry level from the Groningen Growth and Development Centre’s 60 industries databases for 26 OECD countries for the period from 1979 to 2003. The dataset I use has been updated in September 2006. Most variables are taken from the OECD STructural ANalysis (STAN) database, and labour force surveys. The variables which can be found in the database are current value added at national or at basic prices depending on the evaluation used in the national accounts, value added deflators, persons engaged and hours worked. The current value added of each single industry represents the contribution to total GDP.
The labour productivity I calculated as value added per total hours worked, where total hours worked is the product of number of employees and average hours worked. This should reflect changes in average work time per employee, multiple job holding, self employment and the quality of labour.
Lmit = i M (VAi/(Emp*Hours)i) (7)
where Lmt is the Labour productivity for total manufacturing m at time t. VA is value added, Emp the number of employees and Hours total hours worked. As value added for each single industry has been measured in current prices I have to use an aggregate deflator to arrive at constant prices for the total manufacturing sector.9
R&D expenditure
Data on R&D expenditure for total manufacturing from 1979 to 2004 is available at the OECD Analytical Business Enterprise Research and Development (ANBERD) database. The data is available at current prices and has thus to be converted into constant prices. To maintain the same comparability as for labour productivity I use the same aggregate deflators. I hereby assume that the industry specific prices include the input prices of R&D, namely, manpower (R&D scientists, management, technicians, administrative personnel, plants and equipment, materials).
There are many methods to calculate R&D capital, which represents the stock of knowledge. The main practical measurement option put forward (both in empirical analysis and national accounts proposals) has been to approximate the volume of knowledge assets by capitalized R&D expenditures, in which the series of expenditures is formed into stock via the perpetual inventory method (PIM) proposed by Griliches (1973). The essence of the PIM is to form a yearly stock estimate by adding new R&D
expenditure in the year to the existing stock and subtracting ‘depreciation’ or obsolescence of the existing stock. In equation form:
KRt+1
=
(1−
) Kt+
Rt (8)where KRt+1 is the R&D capital stock in year t+l, is a constant (or time-independent) rate of depreciation, Kt is the R&D capital stock in year t; and Rt is the R&D expenditure in year t. The model assumes an exogenous constant depreciation over time and is analogous to forming physical capital stock. Although this method is used in most empirical studies there are some constraints to be considered. Knowledge is very heterogeneous and tacit, which puts the underlying construction and concept of a general knowledge stock in question; neither does it ‘depreciate’ in the same manner as physical capital. Furthermore the use of the PIM imposes a linear accumulation methodology. Concerning the point of a depreciation rate, there are several methods for estimating the private of knowledge depreciation, e.g. by payment of the patent renewal fee, production, cost and profit functions and market evaluations studies. However these methods require a lot of information to which I do not have access to, so I have to rely on rates of previous studies. Bosworth and Jobome (2001) found out by applying market valuation (Tobin’s
q)10 literature that it is 15 per cent per annum. I will apply this as depreciation to calculate R&D capital stock.
10 Tobin's q, is the ratio of the market value of a firm's assets (as measured by the market value of its
Geographical distance between the countries
Estimating distances between countries is not straight forward, as economically some countries are more centralized, while in other countries economic activities are geographically dispersed. To apply a standard rule Keller (2002) measured distances between the capitals “as the crow flies”. I decide to measure the geographical distance between the largest cities of the countries by the size of population, which are not always identical to the capital cities, as the economic importance of the city is relevant rather than the political one. I assume population size to serve as proxy for economic importance, although factors such as the rate of employment and dominant industries might also be of interest should also be considered. This still might not capture the real distance in which the transfer of knowledge takes place but as I do not account for the way of transfer, trade (for which shipping or flight connections might be of interest), FDI or research cooperation, it is a standardized way for all countries in the sample. Furthermore in industrially dispersed countries it is not easy to estimate direct transfers as I measure effects on aggregate sector level.
The distance is measured in kilometres “as the crow flies” and is calculated via the distance tool by www.geobytes.com . Due to their relatively large sizes, I measure for Canada and the USA the distance to the largest cities by population11 which are located the closest, so for the USA for it will be the distance to New York for European countries and the distance to Los Angeles being the largest city on the West Coast to Japan, Korea
and Australia. For Canada it would be Toronto for Europe and Vancouver for the Asian Pacific countries. The distance is measured in thousands of kilometres.12
Social cultural distance
Measuring culture is not straight forward, especially not on national level, as it can be shaped by many different influences such as ethnicity or religion. When it comes to cultural distances on country level one of the most used concepts in economic studies are the dimensions of Hofstede. Subsequent studies have validated the results (from
www.geert-hofstede.com). Hofstede (1980) defined culture as “collective programming of the mind.” Index scores on five dimensions derived from his Value Survey Module (Hofstede 1980, 2001) and subsequent research (Hofstede and Bond 1988) are the most widely cited constructs in the organizational study of national culture. Although the surveys have been carried out in the seventies, the values are assumed to not have changed over time as they are still applied in recent studies. They are as follows:
1. large versus small power distance 2. individualism versus collectivism 3. masculinity versus femininity
4. strong versus weak uncertainty avoidance
The first one measures the extent to which less powerful members of organizations and institutions (like the family) accept and expect that power is distributed unequally, national cultures with high score on power distance accept a more unequal distribution of power. This dimension indicates the importance of hierarchies within organizations. The
second one measures the degree to which individuals are integrated into groups, from loose to more cohesive. National cultures with high scores on that dimension tend to be more individual. The third one refers to the value placed on traditionally male or female values (as understood in most Western cultures). So-called 'masculine' cultures value competitiveness, assertiveness, ambition, and the accumulation of wealth and material possessions, whereas feminine cultures place more value on relationships and quality of life. Japan is considered by Hofstede to be the most "masculine" culture (replaced by Slovakia in a later study), Sweden the most "feminine." Anglo-Saxon cultures are considered moderately masculine. The masculinity vs. femininity issue is fundamental for a range of problem solving. The fourth one, uncertainty avoidance, deals with a society's
tolerance for uncertainty and ambiguity; it indicates to what extent a culture programs its members to feel either uncomfortable or comfortable in unstructured situations. Unstructured situations are novel, unknown, surprising, and different from usual. Uncertainty avoiding cultures try to minimize the possibility of such situations by strict laws and rules, safety and security measures, and on the philosophical and religious level by a belief in absolute truth. National cultures with high scores on that dimension tend to be more uncertainty avoiding. A fifth dimension, long term orientation has been identified especially for East Asian states, but not measured for all countries, so I will leave it out of the analysis. As the weight of the distances within the four dimensions on spillovers is unknown I will apply a measure by Kogut and Singh (1988) to calculate social distance which incorporates all of them weighted by their variance:
CDcg denotes cultural distance; Iig stands for the index for the G5 and Iic for the non-G5 country. Vi is the variance index of the ith dimension. The larger the differences in indices are the more culturally distant are the countries in this index. Division by the variance of this index gives the relative distance of the index in relation with the other indices.13
Cross country immigrant networks
The numbers of immigrants in the source and the host country is used as a proxy for the strength personal networks between two countries and could be interpreted as the total number of potential connections forming social networks between the two countries.
Migration has become an important issue in formulating policies in the OECD countries, especially in the late 1990s, after the fall of the iron curtain. Quality and comparability of data however has not been able to keep pace with the development. Most of the data on cross border movements is collected by national census and labour surveys, which are carried out in different frequencies depending on the country (Lemaitre 2005). Since harmonized data on migration was only collected at the end of the
90’s, I can only measure its effect for a short period. The disadvantage is that I cannot observe how the development of migration flows is related to the convergence of labour productivity over a longer period, and thus cannot control for the long term dynamics in the relationship between migration movements and productivity growth. To incorporate and quantify the impact of links between immigrants in the technology sender country and the countries of origin of the immigrants on cross country R&D spillovers I will use
the number of migrants of the OECD countries in the G5 countries for the years 1995 to 2003. The data can be found in the international migration tables of the OECD at
http://stats.oecd.org/. Four types of sources are used: population registers, residence permits, labour force surveys and censuses. Unfortunately not all countries have data about the origins of their immigrants available. The analysis has hence to be done for smaller samples, when accounting for the effects of migrant networks. I assume that the size of number of migrants increases the possibility for the formation of social networks based on nationality. The non-availability of data for migrants in some of the G5 countries might also signal that their number might be negligible, so that the probability for networking effects on productivity is small.14
One might suggest that geographical and socio-cultural distances are dependent on each other. A Pearson correlation analysis is carried out to discover the strength, shape and the exact direction of the relationship between the country relationship specific variables. Therefore I correlated the geographical distance and cultural distance to the G5 countries. As the number of A value near -1 indicates a strong negative linear relationship; a value near +1 indicates a strong positive linear relationship.
Table 1: Correlation matrix for geographical and socio cultural distance.
D CD
D 1,00
CD ,181 1,00
* Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).
From table 1 above it can be seen that the correlation is 0,181 and not significant between the two parameters, geographical distance and socio-cultural distance. This implies that the two distances are independent of each other; countries which are geographically close to each other do not necessarily share more cultural values.
SECTION 7: DATA ANALYSIS Diagnostic tests
To examine the quality and relationship of the variables I will conduct some diagnostic tests. The data will be tested for nonstationarity, multicollinearity and heteroskedasticity. I note that the following section is based on Hill, Griffiths, Judge (2001).
Nonstationarity:
yi,t = iyit-1 + Xit i + it
where i = 1,2,…, N cross section units or series observed over periods t = 1,2,….,T
Xit represents the exogenous variables in the model, i are the autoregressive coefficients and it the error term. The null hypothesis states that, if | i|=1, then yi contains a unit root. There are several types of unit root tests for panels for which I will not go to deep into detail, but make just a few remarks to explain which tests I chose. The most common unit root tests for panels are the Levin Lin and Chu (LLC 1992), Breitung (2000), Im, Pesaran and Shin (IPS 1997), Hadri (1999) and Fisher’s test for ADF and Philippe Perron (PP), which is proposed by Maddala and Wu (1999). LLC and Breitung assume that there is a common unit root for all cross sections i = for all I, while the others IPS and the Fisher-ADF and Fisher–PP let vary across cross sections, assuming individual unit roots. Maddala and Wu (1999) compared the different tests. The differences are that the Fisher test is based on combining the significance level of the different tests and the IPS is based on combining test statistics. Maddala and Wu therefore used Monte Carlo simulation and came to the conclusion that the Fisher test outperforms the other tests in power.
Table 2: Unit root test for Labour productivity, R&D capital stock in the OECD countries and R&D capital stock in the G5 and the number of migrants.
* Probabilities for Fisher tests are computed using an asymptotic Chi-square distribution. All other tests assume asymptotic normality
It can be seen that at level for all three variables, the null of no unit root cannot be rejected. For the first difference of L and S and the MIG the null can be rejected. A method of stationarizing the data for L and S and MIG would hence be using first differences in the analysis.
Multicollinearity
Multicollinearity exists when independent variables are linear functions of each other. A Pearson correlation analysis is carried out to discover the strength, shape and the exact direction of the relationship between the country relationship specific variables, and to test for multicollinearity. The Pearson correlation is the appropriate one, for variables measured at interval level, while the Kendall and Spearman coefficients assume only measurement at ordinal level. Covariance will be used to linear association between the variables. A value near -1 indicates a strong negative linear relationship; a value near +1 indicates a strong positive linear relationship. When the correlation is larger than 0.8, we
can speak of heavy multicollinearity. Table 3 shows the results. Here I extinguish between domestic R&D capital stock Sd and the R&D capital stock in the G5 countries Sf as these are treated as two variables in the regression analysis.
Table 3: Correlation matrix for domestic and foreign (G5) R&D capital stock and number of migrants in the G5 countries
Sd Sf MIG
Sd 1,00
Sf 0,25** 1,00
MIG 0,40** 0,11 1,00
* Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).
As can be seen, there is some significant correlation between domestic R&D capital stock
Sd and the other two explaining variables. The correlation is not strong however, so that
there are no implications for the estimation procedure.
Heteroskedasticity
result than would be obtained if the error terms were homoskedastic, sometimes lead researchers to reject a null hypothesis that shouldn’t be rejected. The standard errors usually computed for the least squares estimators will be incorrect. Confidence intervals and hypothesis tests that use these standard errors may be misleading (Griffiths, Hill, Judge 2001). I plotted the residuals when doing the regression analysis. If the errors are homoskedastic, there should be no patterns of any sort in the residuals, if they are heteroskedastic, they may tend to exhibit greater variation in some systematic way. When plotting the residuals I could not detect any clear systematic pattern. I assume that the data do not cause problems with heteroskedasticity.
Estimation results and discussion
Table 4: Estimation results Model 1: dln(Lct)= c + t + dln(Sdct)+ ct ; Model 2: dln(Lct)= c + dln(Sdct)+ ct ; Model 3: dln(Lct)= t + dln(Sdct)+ ct ; Model 4: dln(Lct)= c + dln(Sdct) + dln( SfFrancen ct)+ ct; Model 5: dln(Lct)= c + dln(Sdct) + dln( SfGermany ct)+ ct;
Model 1 Model 2 Model 3 Model 4 Model 5
Coeff. Prob Coeff. Prob Coeff. Prob Coeff. Prob Coeff. Prob
Table 5: Estimation results Model 6: dln(Lct)= c + dln(Sdct) + dln( SfJapan ct)+ ct; Model 7: dln(Lct)= c + dln(Sdct) + dln(SfUKct)+ ct Model 8: dln(Lct)= c + dln(Sdct) + dln(SfUSAct)+ ct; Model 9: dln(Lct)= c + dln(Sdct) + dln( Sfgct ) + ct; Model 10: dln(Lct)= c + dln(Sdct) + dln(Sfgct e - CDg)+ ct
Model 6 Model 7 Model 8 Model 9 Model 10
Coeff. Prob. Coeff. Prob. Coeff. Prob. Coeff. Prob. Coeff. Prob.
Table 6: Estimation results
Model 11 Model 12 Model 13 Model 14 Model 15 Model 16 Coeff. Prob Coeff. Prob Coeff. Prob Coeff. Prob Coeff. Prob Coeff. Prob
Table 7: Estimation results
Model 17 Model 18 Coeff. Prob Coeff. Prob
In models 1 to 3, I regress domestic R&D capital stock using cross section fixed (model 2) and time period effects (model 3). Model 2 delivers the best results so far as with an adjusted R-squared of 0.48 which indicates that 48 percent of the variation between the across the countries can be explained by the model. The influence of Sd on L is significant and has a strength of 0.09. The continuing models show the results when applying cross section fixed effects.
uses such an iterative technique to find the values of the parameters in the relationship that cause the sum of squared residuals to be minimized. In some cases the parameters in the exponent cannot be computed. In that case the starting values for cultural distance failed to lead to convergence of the estimator, so that no result could be produced.
controlling for the coefficient in model 18, is strongly negative but not significant. There is hence no direct significant effect of cultural distance. In summary, I do not find distances between the G5 and other countries to play a role on R&D spillover effects; neither do migrants strengthen the spillover effects. However, I find that migration has a slightly negative effect on a country’s productivity. The effect of cultural distance could not be analyzed for R&D spillovers, but I did not find it to have a significant effect when analyzing the impact of migration, as was my intention.
SECTION 8: CONCLUSION
productivity growth. The results would indicate that countries with net emigration rather lose than gain from an international labour market at short term.
Limitations
Availability of data was the largest constraint in the analysis. The limitations of these studies are the small sample size of 19 OECD countries and the time period from 1995 to 2003. Variations in the different levels of labour productivity and R&D investments could hence be due to variations within a business cycle. Most of the countries are in Europe. The R&D capital stock was only measured for aggregate manufacturing. Some industries might exhibit stronger spillover effects than others, I did not account for the size or the weights on spillovers. Spillover effects to service industries have not been considered in this analysis. Furthermore the data for migrants should be more elaborate, accounting for duration and studies or employment of stay, as building a network and the resulting aggregate effects might be significant only after some time lag. The analysis for migration could only be done for small samples at different sizes for each G5 country. Especially data from Korea and Japan, which belongs to the newly industrialized economies, is missing. Many variables have not been taken into account, so that the true effect of migration might be different from the results.
Implications for future research
migrants, as these societies tend to be more cohesive, have stronger links to their roots, and seem to exhibit greater feedback effects on their home countries. One example is the connection between local clustering of high tech industries in Silicon Valley and its link to industrial and research centres in Hsinchu, Taiwan, Singapore or Bangalore, India. Although this thesis only shows results with limited evidence, it should give incentives for further research in the mechanisms of global convergence by using a more
REFERENCES
Audretsch D.B., Feldman M.P. 2003 Knowledge spillovers and the geography of innovation The Handbook of Urban and Regional Economics, Volume 4, Indiana University and Centre for Economic Policy (CEPR), University of Toronto Aldrich H.E., Zimmer C., 1987 Resource mobilization through ethnic networks Sociological Perspectives; October, Vol. 30 Issue 4, pp.422-424
Barba Navaretti G., Venables A.J. 2004 Multinational firms in the world economy Princeton University Press, Princeton N.J. 08540, U.S.A.
Beine M., Docquier F., Rapoport H. 2001 Brain drain and economic growth: Theory and evidence Journal of Development Economics, Vol.64, pp. 275-289
Bosworth D., Jobome G. 2001 The rate of depreciation of technological
knowledge: Evidence from patent renewal data, Economic Issues, vol. 8, part 1, pp. 59–82,
Branstetter L. 2000 Looking for international knowledge spillovers: A review of the literature with suggestions for new approaches from “The Economics and Econometrics of Innovation” 495-518, Encaoua D, Hall B.H., Laisney F., Mairesse J., Kluwer Academic Publishers, the Netherlands
Branstetter L. 2002 Measuring the link between academic science and innovation: The case of California research universities, mimeo, University of California, Davis
Breitung, J. 2000 The local power of some unit root tests for panel data. In B. Baltagi
(Ed.), Advances in econometrics, vol. 15: Non-stationary panels, panel cointegration, and dynamic panels, (pp. 161–78) Amsterdam: JAI Press.
Brüderl, J. 2005 Panel Data Analysis, University of Mannheim, March, http://www.sowi.uni-mannheim.de/lehrstuehle/lessm/veranst/Panelanalyse.pdf
Coe D.T., Helpman E.1991 Innovation and growth in the global economy MIT Press, Cambridge, MA, USA
Coe D.T., Helpman E.1995 International R&D spillovers European Economic Review 39(5), 859-887
Coe D.T., Helpman E., Hoffmaister A.W., 1997 North-South R&D spillovers The Economic Journal, Vol. 107, No. 440, pp.134-149
Cohen W.M., Klepper S. 1991 Firm size versus diversity in the achievement of
Docquier F., Lodigiani E. 2006 Skilled migration and business networks Centro Studi Luca d’Agliano Development Studies Working Papers
Feldman M.P. 2002 The internet revolution and the geography of innovation International Social Science Journal 54:47-56
Ferri G., Serlenga L. 2006 Ethnic diversity and trade Working paper, University of Bari Griliches, Z. 1973 Research expenditures and growth accounting William, Z.
(ed.), Science and Technology in Economic Growth, Macmillian, London, pp. 59–95.
Griliches Z. 1990 Patent statistics as economic indicators: A survey Journal of Economic Literature, Vol. 28, December, pp. 1661-1707
Groningen Growth and development center:http://www.ggdc.net/index-dseries.html
Version September 2006
Guellec D., Pottelsberghe de la Potterie B. 2001 R&D and productivity growth: Panel data analysis of 16 OECD countries OECD Economic Studies, Hill R, Griffiths W. E, Hall B. 1993 Has the rate of return declined? The Brookings paper on Economic
Activity, vol. 2: Microeconomics
Hofstede G.H. 1980 The cultures consequences: International differences in work-related values SAGE, California. USA
Hofstede, G. H., M. H. Bond 1988 The Confucius connection: From cultural roots to
economic growth Organizational Dynamics, Spring, 5-21.
Hofstede G.H. 2001 Culture's consequences, Second edition: Comparing values, behaviors, institutions and organizations across nations SAGE, California. USA Hsu J.Y, Saxenian A.L. 2000 The limits of Guanxi Capitalism: Transnational
collaboration between Taiwan and the USA Environment and Planning A, November 32, issue 11, pp. 1991-2005
Im, K. S., Pesaran, M. H., Shin, Y. 2003 Testing for unit roots in heterogeneous panels
Journal of Econometrics, 115, pp. 53–74
Jacobs J. 1969 The economy of cities Random House, New York
Jaffe A.M., Traijtenberg M, 1996 Flows of knowledge spillovers Working paper, Strasbourg Conference on the Economics and Econometrics of Innovation
Judge G. G. 2001 Undergraduate Econometrics 2nd Edition, John Wiley & Sons, Inc. Keller W. 2002 Geographic localization of international technology diffusion Working paper, NBER, University of Texas
Kerr W.R. 2007 Ethnic scientific communities and international technology diffusion Working paper, Harvard Business School and MIT
Kogut B., Singh H. 1988 The effect of national culture on the choice of entry mode Journal of International Business Studies, Vol. 19, No.3 pp.422-432
Krugman P.1987 The narrow moving band, the Dutch disease and the competitive consequences of Mrs. Thatcher: Notes on trade in the presence of dynamics scale of economies Journal of Development Economics, 27, pp. 41-55
Krugman P. 1991 Geography and trade MIT Press, Cambridge
Kugler M., Rapoport H. 2005 Skilled emigration, business networks and foreign direct investment CESIFO working paper no.1455
Lemaitre G. 2005 The comparability of international migration statistic: Problems and prospects OECD working paper
Levin A, Lin C, Chu J. 2002 Unit root tests in panel data: asymptotic and finite properties Journal of Economics 1008 (2002) 1-24
Madalla G.S. 1999 On the use of panel data methods with cross-country data Annales D’Économie et de Statistque.-No 55-56
Maddala G.S., Wu S. 1999 A comparative study of unit root tests with panel data and a new simple test Oxford Bulletin of Economics and Statistics, Special issue 0305-9049 Madden G. Savage S.J. 2000 R&D spillovers, information technology and
telecommunications, and productivity in Asia and the OECD Elsevier Information Economics and Policy 12, pp.367-392
Manski C.F. 2000 Economic analysis of social interactions Journal of Economic Perspectives, 14: pp. 115-136
Malmberg A. 1997 Industrial geography: Location and learning Article, Progress in Human Geography, Sage Publications
Mody A. 2002 Is FDI integrating the world economy? International Monetary Fund, Washington D.C. processed
Nabeshima K. 2002 Quality of secondary education in East Asia World Bank, Washington D.C., processed
Nelson, R.R., 1990 What is public and what is private about technology? Working Paper No 90-9, University of California Center for Research in Management,
OECD 2005/2006 Research and development expenditure in industry ANBERD 1987-2004
OECD 2008 A profile of immigrant populations in the 21st century: Data from OECD
countries
Pedroni, P. 1999 Critical values for cointegration tests in heterogeneous panels with multiple repressors’, Oxford Bulletin of Economics and Statistics, special issue, 61, pp 653-670.
Pottelsberghe de la Potterie, Lichtenberg F. 2001 Does foreign direct investment transfer technology across borders? Review of Economics and Statistics 83: 490-497
Rauch J.E., Trindade V. 2002 Ethnic Chinese networks in international trade Review of Economics and Statistics 84(1):116-30
Romer P. 1990 Endogenous technical change Journal of political economy, 98 S71-S102
Saxenian A. 1993 Regional advantage: Culture and competition in Silicon Valley and Route 128 Harvard University press, Cambridge, MA, USA
Schreyer, P. 2000 The contribution of Information and Communication Technology to output growth: A Study of the G7 Countries STI Working Papers 2000/2, OECD, Paris
Schreyer, P. 2002 Computer price indices and international growth and productivity comparisons Review of Income and Wealth, March 48 (1), pp. 15-31,.
Stoneman, P., 1983. The Economic Analysis of Technological Change , Oxford University Press, Oxford, UK
Tong S.Y. 2005 Ethnic networks in FDI and the impact of institutional development Review of development economics, 9 (4), pp. 563-580
Venables J. 2001 Geography and international inequalities: The impact of new technologies Working paper Centre’s Globalisation Programme
Von Hippel E. 1994 Sticky information and the locus of problem solving: Implications for innovation Management Science 40: pp. 429-439
Yaffee, R.A. 2003 A primer for panel data analysis University of New York,
Information Technology Services
Young A. 1991 Learning by doing and the dynamic effects of international trade Quarterly Journal of Economics, 106, pp. 369-406
Yusuf S. 2003 Innovative East Asia: The future of growth The World bank, Oxford University Press, New York, NY 10016
www.geert-hofstede.com www.geobytes.com
APPENDIX
A: Calculation of labour productivity at constant prices
I note that these procedures based on Schreyer (2000; 2002).
If PiV is the deflator for value added in industry i, PV the aggregate deflator, then the deflator change in period t is given by:
PV,t = i siV,t PiV,t.
The price change in industry i is weighted by the average share s of industry i in total manufacturing value added over the two periods and defined as:
siV,t = ½ (PiV,t Vit/ i(PiV,t Vit) + PiV,t-1Vit-1/ i(PiV,t-1 Vit-1))
The real aggregate value added growth is then calculated as the growth rate of aggregate current value added minus the growth rate of the deflator. As follows:
lnVt = ln Vt PV,t – ln PV,t
The deflated labour productivity dLmt is then calculated as:
dLmt = Lmt / (1+ lnVt)
B: Data Table 1:
Labor productivity for total
Table 2:
R&D capital stock in Mio. PPP -
US$ Constant prices 1995 1996 1997 1998 1999 2000 2001 2002 2003
Table 3: Geographic distances to the G5 counties
Distance in 1000 km France Germany Japan UK USA
Australia 16,978 16,499 7,835 17,009 12,081 Belgium 0,262 0,175 9,461 0,318 5,889 Canada 6,007 6,339 7,555 5,72 0,703 Czech Republic 0,886 0,28 9,079 1,035 6,576 Denmark 1,028 0,288 8,699 0,954 6,191 Finland 1,91 1,165 7,827 1,82 6,619 France 0,479 9,723 0,343 5,838 Germany 0,479 8,888 0,719 6,204 Ireland 0,779 1,075 9,596 0,463 5,116 Italy 0,64 0,349 9,726 0,96 6,467 Japan 9,723 8,888 9,567 8,814 Korea 8,975 8,558 1,158 8,864 9,591 Netherlands 0,429 0,179 9,300 0,356 5,866 Norway 1,343 0,71 8,414 1,153 5,916 Poland 1,367 0,516 8,588 1,447 6,854 Spain 1,054 1,486 10,775 1,267 5,772 Sweden 1,545 0,81 8,179 1,431 6,32 UK 0,343 0,719 9,567 5,572 USA 5,838 6,204 8,814 5,572
Table 4: National cultural differences to the G5 countries by scores according to Hofstede
Cultural Distances France Germany Japan UK USA
Table 5: Number of migrants in G5 countries in thousands
Country of origin Year In France In Germany In Japan In the UK In the USA
