WPS4144 Trade and Human Capital Accumulation - Evidence from U.S. Immigrants Dörte Dömeland* Abstract This study provides empirical evidence that trade increases on-the-job human capital accumulation by estimating the effect of home country openness on estimated returns to home country experience of U.S. immigrants. The positive effect of trade on on-the-job human capital accumulation remains significant when controlling for GDP, educational attainment and institutional quality. It is not the result of self-selection, heterogeneity in returns to experience, English speaking origin or cultural background. The effect persists when restricting the sample to non-OECD countries, thereby resolving the theoretical ambiguity whether trade increases or decreases learning-by-doing. The role of trade in generating economic growth is therefore likely to be more important than generally considered. JEL Classiffication: O15, F00 World Bank Policy Research Working Paper 4144, March 2007 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. *I am very grateful to Antonio Ciccone for excellent advice and suggestions and thank Jaume Garcia, Christian Haefke, Adriana Kugler, Martin Menner, Joel Shapiro, Ernesto Villanueva, Lutz Weinke and Michael Wolf for very useful comments. Email: ddomelandnarvaez@worldbank.org. Tel: ++1-202-458-1238 "Human capital accumulation takes place in school, in research organi- zations and in the course of producing goods and engaging in trade." Robert Lucas (1993) 1 Introduction Theory identi...es the e¤ect of trade on on-the-job human capital accumula- tion as an important channel through which trade may enhance economic growth. But whether trade increases or decreases on-the-job human capital accumulation is ambiguous from a theoretical point of view. Trade may fos- ter the acquisition of human capital by facilitating the transfer of ideas and technology from technologically more advanced countries to less advanced economies. Technology transfer a¤ects human capital accumulation through two channels. First, implementing and working with a new technology in- creases the knowledge of workers. On-the-job learning is hence a by-product of trade. Second, trade may lead to an increase in wages of skilled relative to unskilled workers inducing workers to invest more in human capital (e.g. Hall and Jones (1999), Pissarides (1997) and Goh and Olivier (2002)). Opening up to trade may also theoretically reduce on-the-job human capital accumulation. If some productive activities carry a higher rate of skill acquisition than others, moving from autarchy to free trade may de- press learning-by-doing. This happens if trade induces countries to import high-quality goods rather than to produce them (Stokey (1991) and Young (1991)).1 Approaching the question whether trade increases of decreases on-the- job human capital accumulation empirically requires a measure of on-the-job accumulation of human capital, such as the return to experience. Cross- country data on returns to experience is not well suited to estimate the e¤ect of trade on human capital accumulation since the return to experience is determined by the price of on-the-job human capital, as well as its quantity and quality. The identi...cation of cross-country di¤erences in human capital accumulation requires to hold this price constant across countries. But the price of on-the-job human capital is a function of country-speci...c variables such as technology, supply of human capital, labor market institutions and governmental quality. If all these country-speci...c variables were constant across time, panel data would be a solution to this problem. But panel data on cross-country 1Routine or traditional production techniques, for example, are likely to be associated with less learning than more complex, technology-intensive tasks (Lucas (1988, 1993)). 1 returns to experience is generally not available. Moreover panel data could not solve another fundamental problem. Opening up to trade does not only change commodity prices, it also a¤ects relative factor prices and hence the price of human capital.2 As a consequence, observing higher returns to experience in more open economies does not allow to conclude that trade increases on-the-job human capital accumulation. These problems are likely to explain why there exist very few empirical studies about cross-country di¤erences in on-the-job human capital accu- mulation. The empirical strategy pursued in this paper overcomes these issues as it does not rely on cross-country data. Instead, it uses data on US immigrants from di¤erent source countries to estimate US returns to home country experience. These returns are measured within the same labor mar- ket and are therefore not a¤ected by cross-country di¤erences in the price of on-the-job human capital. Furthermore, the US labor market is character- ized by a relatively low level of labor market regulation which assures that wages and hence estimated returns to experience are related to productivity. I provide evidence of a positive and signi...cant e¤ect of home country trade on returns to home country experience of US immigrants, indicating that trade enhances the accumulation of on-the-job human capital. High investment rates per worker, which can be considered a precondition for technology adoption, and strong governments and institutions have a pos- itive and signi...cant e¤ect on returns to home country experience. The positive e¤ect of trade on returns to home country experience persists when restricting the sample to developing countries, providing empirical evidence for the hypothesis that trade increases learning-by-doing through technol- ogy transfer even in less developed countries. It is robust to the issues of self-selection, heterogeneity in returns to experience and English speaking origin and is unlikely to be the result of unobserved cultural background. The remaining part of this paper is organized as follows. The next section provides a short summary of the related literature. Section 3 summarizes the empirical strategy. Section 4 discusses data and provides a preliminary data analysis. The main results are presented and discussed in section 5. Section 6 performs a series of robustness checks. The ...nal section concludes. 2According to the Stolper-Samuelson theorem the relative reward of a factor that is more intensively used in the production of a good increases as the price of the good rises. 2 2 Related Literature Human capital accumulation on the job, either through learning by doing or formal training does not come for free. Individuals have to invest time, e¤ort and monetary costs in terms of direct training costs and foregone earnings in order to increase their human capital. How does trade inuence this individual investment decision? A standard model which provides basic insights into the human capital investment decision is the Ben Porath Model (Ben Porath (1967)). This model predicts that the optimal amount of human capital investment de- pends on the wage rate per unit of human capital, the cost of training and the rate of obsolence. Bartel and Sicherman (1998) show that in the Ben Porath Model an increase in the wage rate per unit of human capital un- ambiguously increases investment in each period. Raising the productivity of human capital may reduce the cost of training and/or increase the value of time in training relative to work. Both changes enhance investment in training. On the other hand, the introduction of new work processes may make existing human capital obsolete. A higher rate of obsolence of human capital may lower investment in human capital. If opening up to trade induces countries to specialize in the production of goods whose production technology carries a low rate of learning, then human capital accumulation is likely to decrease. But if trade leads to technology transfer then opening up to trade may speed up learning in technologically less advanced countries because implementing and working with the new technology increases the knowledge of workers. This positive e¤ect of trade on human capital accumulation is reinforced if the transferred technology is skill biased by raising the demand for skilled workers relative to unskilled workers permanently. A higher relative demand for skilled workers leads to an increase in the wage rate per unit of human capital, triggering human capital investment. At the same time the transfer of skill-biased technologies may depress investment in human capital if, for example, the introduction of new products or production technologies makes existing skills obsolete at a faster rate.3 Consistent with the idea that trade increases the demand for skilled labor, there exists empirical evidence of a positive association between trade and relative wages of skilled workers. Using cross-country data, Denny, 3The transfer of new technologies may not only a¤ect the optimal amount of human capital investment, but also the types of skills which workers would like to acquire. Workers may, for example, be more likely to invest in the accumulation of skills that are more highly valued at the world technology frontier. 3 Harmon and Lydon (2001) identify higher returns to education in more open countries. Several country studies that analyze the e¤ect of trade on relative wages of skilled and unskilled workers conclude that technology transfer is the most likely reason for the increase in wage inequality (e.g. Robbins (1994, 1995), Hanson and Harrison (1994)). Still, observing higher relative wages in more open economies does not allow to conclude that trade enhances human capital accumulation since relative wages are likely to be a¤ected by country speci...c factors, such as technology, supply of human capital or labor market regulations. Direct evidence on the relation between trade and human capital accumulation is far from conclusive. Using a cointegration analysis, Chuang (2000) ...nds a bidirectional Granger causality between exports and the share of individuals who have attained higher education in Taiwan. Alcala and Ciccone (2001) conclude that trade increases productivity in the cross-country context, but ...nd no statistically signi...cant association between openness and the average level of human capital, as measured by years of education. None of these studies analyzes the e¤ect of trade on on-the-job human capital accumula- tion. This is not the ...rst study using data on US immigrants in order to deduce information about their country of origin. Hanushek and Kim (1999) and Bratsberg and Terrel (2002), for example, analyze the e¤ect of home country school quality on earnings of US immigrants. Hanushek and Kim (1999) ...nd a strong and positive e¤ect of international math and science test scores on earnings and returns to education of US immigrants. Bratsberg and Terrel (2002) conclude that holding per-capita GDP constant, immigrants from countries with lower pupil-teacher ratios and greater expenditures per pupil earn higher returns to education in the US. Borjas (1998) provides empirical evidence about the e¤ect of source country characteristics on US immigrant quality. This is the only study to my knowledge that relates openness to earnings of US immigrants. He ...nds no signi...cant e¤ect of openness on the log entry wage of US immigrants when controlling for country ...xed e¤ect and/or educational attainment. None of these studies analyzes the e¤ect of home country characteristics on on-the-job human capital accumulation of US immigrants. 4 3 Estimating the e¤ect of trade on returns to ex- perience This paper addresses the question whether trade increases or decreases on- the-job human capital accumulation by relating US returns to home country experience of immigrants to home country openness. The methodology used is similar to the two-step procedure proposed by Card and Krueger (1992). The ...rst step consists of estimating a Mincerian earnings equation to obtain estimates of country-speci...c returns to home country experience. In the sec- ond step, these estimated returns to home country experience are regressed on a measure of home-country openness and other control variables. In the ...rst step, a Mincerian earnings equation is estimated for each country separately. This equation relates log of earnings to years of school- ing, potential labor market experience and its square. Labor market expe- rience of immigrant i from country j can be decomposed into pre-migration experience (Hij) and post-migration experience (Uij). Assuming that the ef- fect of experience in a country is linear in experience and its square, the earn- ings function of immigrants who completed their education in their source country can be written as ln yij = 2 j + U2 + . (1) jSij + 1jHij + 2jHij + 1jUij + 2j ij ij The return to home country experience equals @ ln yij= + 2 H. @Hij 1j 2j 1j is the slope of the log earnings experience pro...le. 2j captures the curvature and is usually negative, leading to the familiar concave experience earnings pro...le. H is the number of years at which returns to home country experience are evaluated. The coe˘ cients in equation (1) are not held constant across countries since there exists convincing evidence that the intercept as well as returns to individual characteristics of US immigrants, such as education or time spent in the US, are likely to vary across countries of origin. Borjas (1998), for example shows, that the log entry wage varies with source country char- acteristics. Hanushek and Kim (1999) and Bratsberg and Ragan (2002) provide evidence that returns to education vary substantially across coun- tries, reecting both di¤erences in quality of education and transferability of skills. The question whether trade increases or decreases on-the-job human capital accumulation is addressed in this study by relating home country 5 openness of US immigrants to their returns to home country experience. Assuming a linear relation between openness and return to experience, this question translates into 1j + 2 2jH = aH + bHOpenj: (2) where Openj is some measure of openness in country j. A positive (negative) sign of the estimator of the coe˘ cient bH indicates that trade increases (decreases) on-the-job human capital accumulation. The marginal e¤ect of openness on returns to home country experience is restricted to be constant across countries in speci...cation (2): This assumption will be relaxed during the empirical analysis for some speci...cations and bH will be allowed to vary for sub-groups of countries.4 Relying on the fact that the return to experience captures productive capabilities attributable to on-the-job human capital investment, this es- timation strategy requires to exclude that factors that are not related to on-the-job human capital investment may generate the upward slope of the experience earnings pro...le at the beginning of a worker's career. In a search environment, for example, wages may grow with labor market experience be- cause workers may improve the quality of their jobs by means of job search (see for example, Jovanovich (1979)). Moving to the United States is likely to imply for immigrants that the search capital is lost. Employer schemes to economize on costs of monitoring (Lazear (1981)) and turnover costs (Salop and Salop (1976)) may also generate an upward sloping experience earnings pro...le but are unlikely to explain positive returns to home country experi- ence of US immigrants.5 It is therefore reasonable to assume that returns to home country experience of US immigrants are related to on-the-job human capital investment. To obtain an estimate of the return to home country experience, I sub- stitute potential experience Eij Hij + Uij for home country experience Hij; 4Openness does not only vary across countries but also over time t, that is 1jt+ 2 2jtH = ajH + gtH + bHOpenjt: (3) Identifying the home country ...xed e¤ect (ajH) and the time e¤ect (gtH) requires to construct a panel data set on returns to experience. This is severely constrained by the limited amount of observations available for a large number of countries in the US Censuses. 5In these models wages grow because ...rms defer compensation in order to prevent workers from shirking (Lazear (1981)) or in order to induce a self-selection of heterogeneous workers that enhances productivity (Salop and Salop (1976)). 6 ln yij = 2 j+ (4) jSij + 1jEij + 2jEij + ( + 2 (2 1j 1j )Uij + (2j 2j)Uij 2j)EijUij + ij where lnyij is the log of annual earnings for immigrant i from source country j: Sij is a series of dummy variables for di¤erent degrees of schooling. Uij is measured as the di¤erence between the census year and the year at the midpoint of the year of immigration bracket.6 As can be seen from equation (4), the coe˘ cient on Uij consists of the di¤erence between the return to post-migration and pre-migration experience. Potential experience Eij is de...ned as age minus years of education minus six. The second step of the two-step procedure consists of regressing the estimated returns to experience on a measure of openness, that is Xj is a set of observed country-speci...c characteristics, such as GDP c 1j + 2cH = 2j H + (5) H Openj + H Xj + uj per capita, investment per worker, average years of education in the home country and governmental quality. These variables will be discussed in detail when presenting the results. Unobserved country-speci...c characteristics are captured by uj. Under the assumption that openness a¤ects only the slope of the earnings- experience pro...le but not its curvature, the e˘ ciency of the estimator of could be improved by estimating the e¤ect of openness in one step. The H two-step procedure, however, has several important advantages. First, it provides a straight-forward interpretation of the results by allowing to esti- mate the e¤ect of openness on returns to home country experience and not only on the slope of the earnings-experience pro...le. Second, being compu- tationally less burdensome it facilitates the estimation of extremely exible forms of the ...rst stage regression. In one of the speci...cations presented below returns to experience are, for example, allowed to vary with level of education. Third, it allows to illustrate the diversity in returns to experience across countries. A special case of (2) is to assume that with the exception of 1j the coe˘ cients on the Mincerian equation do not vary across countries. This implies that openness can only a¤ect the slope of the experience-earnings pro...le but not its curvature. 6 Data on US immigrants is taken from the 1980 and 1990 US Censuses. 7 Under this assumption, equation (4) may be written as ln yij = + Sij + a0Eij + b0EijOpenj + 2Eij 2 2 E U 2 ij ij (6) +( 1 a0)U b0UijOpenj + ij This equation forms the starting point of my empirical analysis. The results will be presented in the following section. 4 Data Analysis My empirical analysis uses data from the 1980 and 1990 US Censuses.7 The dependent variable of the Mincerian earnings equation is the natural logarithm of the annual wage or salary income in the year preceding the census. The set of control variables includes potential experience, dummies for each year of schooling, years spent in the US and its square, married with spouse present and dummy variables indicating whether the respon- dent speaks only English or speaks English very well, health limiting work, residence in SMSA, eight census divisions and year of immigration. To con- trol for changes in labor market conditions a dummy indicating Census year 1980 is added to the regression and interacted with regional and educa- tional dummies. Interaction terms with the Census year dummy and other explanatory variables are not statistically signi...cant and are therefore ex- cluded from the speci...cations. Descriptive statistics are provided in Table 1.8 The summary statistics of trade used in this analysis is the natural log- arithm of the mean of Open, where the mean is calculated from 1970 to 1980. I will refer to this measure as log Open.9 Using the logarithm of this summary measure of trade implies that the e¤ect of a one percentage point increase in Open on the dependent variable is larger the lower the level of openness. Table 2 presents the results of the Mincerian earnings equation for the two censuses. As can be seen in the last column of Table 2; which combines the 1980 and 1990 Census, the estimated return to home country experience 7 The data are available online at http://www.ipums.org. For more information on the data, see Ruggles and Sobek (1997). 8 A detailed description of the data can be found in the data appendix. 9 Open is de...ned as the ratio of imports plus exports in exchange rate US$ relative to GDP in purchasing-power-parity US$. Other studies that use this measure are Alcala and Ciccone (2001) and Dollar and Kraay (2002). 8 evaluated at ...ve year experience is 1.8 percent a year and declines to 1.4 percent when evaluated at ten years of experience. It is a well documented fact that returns to home country experience of immigrants are on average low compared to returns to experience of native born US citizens. Using data from the 1970 Census, Chiswick (1978), for example, reports returns to experience of 1.4 percent for immigrants and 2.1 percent for US natives. Adding openness to the regression and estimating equation (6) leads to the estimates presented in Table 3. The estimated return to home country experience amounts to 2.1 percent evaluated at 5 years of labor market experience and the mean of log Open. Imposing the constraints on b0 and 2 implied by equation (6) reduces the estimated return to home country experience to 1.6 percent. The coe˘ cient on the interaction term between log Open and experience has a positive sign and is signi...cant.10 Increasing Open from 0.2 to 0.3 raises the return to home country experience by 0.2 percentage points if coe˘ cients are unconstrained.11 Taking a country from the 10th percentile to the 90th percentile of log Open raises the estimated return to home country experience by 1.2 percentage points and increases to 1.3 percentage points in the constrained regression. Based on the results, Figure 1 displays the predicted log earnings-experience pro...les for two US immigrants. Both immigrants are assumed to have the same individual characteristics, but the source countries of the immigrants di¤er in their degree of openness. As equation (6) imposes that neither the intercept of the Mincerian earnings equation nor the return to individuals characteristics vary with country of origin, the intercept of the log earnings- experience pro...le is the same for both immigrants. Moreover, since equation (6) assumes that openness a¤ects only the slope but not the curvature of the pro...le, the di¤erence in returns to experience among the two immigrants remains constant throughout their work life. Restricting the intercept of the Mincerian earnings equation to be con- stant across countries may lead to a biased estimate of the e¤ect of openness on the slope of the log earnings-experience pro...le.12 If initial wages of US immigrants from more open economies are higher, then the e¤ect of open- 10The calculation of the standard errors takes into account heteroscedasticity and clus- tering. Not controlling for clustering may lead to a serious downward bias in the OLS standard errors when adding aggregate market variable to micro units. (see Moulton (1986, 1990)) 11The change in Open from 0.2 to 0.3 corresponds approximately to an increase from the 20th percentile to the median value. To give an example, this change corresponds to South Korea (Open equals 0.20) as compared to Taiwan (0.32) or Australia (0.20) as compared to Canada (0.33). 12It may also lead to a biased curvature of the pro...le. 9 ness on returns to experience is likely to be overstated.13 Controlling for country ...xed e¤ects in the last column of Table 3 allows for cross-country variation in the intercept and leads to an estimated return to home country experience of 1.9 percent evaluated at 5 years of labor market experience and the mean of log Open. When controlling for country ...xed e¤ects, the coe˘ cient on the interaction term between experience and openness remains signi...cant, but drops from 0.006 to 0.002. This implies that raising Open from 0.2 to 0.3 leads to an increase in the annual return to experience by 1 percentage point. As pointed out above empirical evidence suggests that not only the inter- cept of the Mincerian earnings equation, but also the returns to individual characteristics, such as education or post-migration experience, are likely to vary across source countries of US immigrants. This issue can be addressed by estimating the e¤ect of openness on returns to experience by means of the two-step method. 5 Openness and Returns to Experience The ...rst step of the two-step method consists in estimating the Mincerian earnings equation for each country separately, yielding estimates of the country-speci...c returns to home country experience.14 Di¤erences in re- turns to home country experience across countries of origin of US immi- grants are substantial. Evaluated at 5 (10) years of labor market experience statistically signi...cant returns range from 7.8 (6.6) percent for immigrants proceeding from Norway, Finland and Japan to 1.4 (1) percent for Philip- pines, Mexicans and Guatemalans (see Table 4). The relatively low average return to home country experience of US immigrants identi...ed in previous 13Immigrants from more open economies may not only have higher returns to experience, but also higher entry wages. As pointed out above, Borjas (1998) ...nds no signi...cant e¤ect of openness on the entry wage level of US immigrants as soon as he controls for country ...xed e¤ects and/or initial educational attainment. However, he ...nds a positive and signi...cant e¤ect of GDP per capita on log entry wages. Given that GDP per capita and openness are positively correlated, trade is likely to a¤ect log entry wages as long as GDP per capita is not controlled for. 14Returns to experience are only estimated for countries with at least 50 US immigrants that satisfy the election criteria, using data from the 1980 and 1990 census jointly. An alternative strategy would have been to estimate country-speci...c returns to experience per census year. However, variation in returns to experience across census years is not signi...cant, inducing me to stack the 1980 and 1990 census. This increases the number of observations used for estimating the country-speci...c returns to experience and therefore the precision of the estimates. 10 studies is henceforth likely to be determined by the fact that immigrants from Mexico form a large share of the overall US immigrant population. Nearly 32 percent of immigrants in my sample are Mexicans. Taking the unweighted average across the 93 countries in the sample leads to a return to home country experience of 3.1 percent with a standard deviation of 3.7. Returns to home country experience change with years of experience. If a country with a higher log earnings experience slope 1j has a larger coe˘ - cient on the curvature of the log earnings experience pro...le in absolute value then the ranking in returns to experience of two countries may be reversed when evaluating returns at di¤erent years of experience.15 Illustrating this fact ...gure 2 shows that the di¤erence between returns to experience of im- migrants from Korea and Taiwan, for example, increases with years of home country experience, while the contrary holds for Belgium and Portugal. For the analysis below, I use ...ve years of labor market experience as the year at which to evaluate returns to home country experience experience. 16 The second step of the estimation strategy consists of regressing esti- mated returns to home country experience of US immigrants on home coun- try openness in order to understand how trade relates to on-the-job accu- mulation of human capital. Table 5 presents the results for three di¤erent samples: all countries, non-Oil countries and non-OECD/non-Oil countries. The dependent variable is the estimated return to home country experience evaluated at ...ve years of labor market experience in percent. As in section 4, the statistic of openness is the natural logarith of the mean of Open, where the mean is calculated from 1970 to 1980 in order to control for short-term uctuation.17 Regressing estimated returns to home country experience on openness yields a positive and signi...cant coe˘ cient, as shown in speci...cation (I) of Table 5: The coe˘ cient amounts to 1.290 when using the entire sample and reduces slightly, when dropping the oil exporting countries. Increasing Open from 0.2 to 0.3 raises the annual return to home country experience by 0.5 percentage points in the non-Oil sample. Given the low returns to home country experience of US immigrants, this e¤ect is quite large. Restricting the sample to non-OECD countries depresses the e¤ect of openness on returns to experience. Still it remains positive and signi...cant.18 15This may arise, for example, if the introduction of new technologies in a country induces young workers to invest more in on-the-job learning, but makes knowledge of older workers obsolete at a faster rate. 16Results for other years of experience are available upon request. 17A detailed description of variables can be found in the Data Appendix. 18Standard errors are calculated using the White estimator in order to correct for het- 11 Controlling for regional dummies in speci...cation (III) approximately dou- bles the R-square, but does not alter the sign and the signi...cance level of the coe˘ cients on log Open.19 These ...ndings indicate a positive and signi...cant e¤ect of trade on returns to home country experience of US immigrants. They do not imply that US immigrants proceeding from more open countries receive more training in their home country. Human capital is not completely transferable across countries. Some part of it evaporates as immigrants cross the border since an immigrant may bring skills which are not marketable in the US. A share of human capital accumulated on the job is ...rm-speci...c and many general skills are tied to a particular product market or technology. As a consequence, higher returns to home country experience do not necessarily indicate that an immigrant accumulated a higher quantity of human capital during his working life. But they show that he accumulated more skills that are valued by the US labor market. This interpretation of returns to experience is not inconsistent with the story underlying the theoretical link between openness and human capital accumulation, arguing that immigrants accumulate skills because they are in contact with production technologies developed in countries which are closer to the world technology frontier. But it is exactly skills related to more advanced technologies that are likely to be valued by the US labor market. Confusion about the causal e¤ect of trade on returns to experience can arise from omitting variables that are correlated with both returns to expe- rience and openness. As long as neglected elements are ...xed within regions this is of no concern as di¤erences in unobservables are absorbed by regional dummies. But if a positive correlation of unmeasured determinants of re- turns to experience and openness persists even after controlling for regions, the estimated coe˘ cient on openness does not reveal the causal e¤ect of openness on returns to experience. This issue can be solved by including country-speci...c variables in the set of explanatory variables that a¤ect re- turns to experience and are correlated with log Open. Prime candidates are GDP per capita and average years and quality of schooling. The correlation coe˘ cient between estimated returns to home country eroscedasticity. Disturbances are likely to be heteroscedastic as the dependent variable in the second stage regression is itself an estimated regression coe˘ cient. 19To control for region-speci...c e¤ects, I keep regional dummies in the speci...cation if they are at least statistical signi...cant at the 10-percent level. There is one exception to this rule. If the deletion of a statistical insigni...cant regional dummy increases the Akaike criterion, the regional dummy is kept. 12 experience and log of real GDP in the entire sample is 0.295.20 This pos- itive association between log of real GDP and returns to experience is not surprising. Resources devoted to schooling and training tend to be higher in richer countries. But even after controlling for school and training resources a signi...cant positive e¤ect of GDP per capita on returns to experience is likely to persist, as countries with a higher GDP per capita are more likely to have production technologies similar to the US. This implies that the types of skills immigrants from richer countries learn on the job are more valued by the US labor market than skills obtained in less developed countries. At the same time a large empirical literature provides evidence of a posi- tive association between GDP per capita and trade. The positive correlation coe˘ cient of 0.475 between log of real GDP and log Open in the data is con- sistent with these ...ndings and indicates that not controlling for GDP per capita is likely to lead to an overestimation of the e¤ect of openness on returns to experience. As expected, regressing returns to home country experience on GDP per capita yields a signi...cantly positive coe˘ cient on GDP per capita and reduces the coe˘ cient on log Open in all three samples. Speci...cation (IV ) in Table 5 reveals that once GDP per capita and regional dummies are controlled for, the coe˘ cient on log Open falls to approximately 0.830, but remains signi...cant at the 5 percent signi...cance level in all three samples. Average years of schooling as well as quality of schooling are likely to be positively associated with returns to experience. According to the Ben- Porath Model, individuals with more schooling and better schooling tend to invest more in human capital on the job. This arises from the fact that individuals with a higher level of education are likely to be more able and/or face lower discount rates. Higher average years of schooling may further re- ect a higher demand for human capital, increasing both the pro...tability of investing in schooling and on-the-job training. A higher level and better quality of average schooling may also lead to lower costs of post-school in- vestment and/or increase its bene...ts for two reasons. First, in the presence of human capital externalities working hand in hand with skilled workers may enhance the individual's learning on the job. Second, a higher level of average education is likely to be positively correlated with the level of human capital of the training sta¤. A higher average level and better quality of schooling is also likely to be positively correlated with openness since it may increase the demand for 20Consistent with log Open, the log of real GDP per capita is de...ned as the log of its mean calculated from 1970 to 1980. 13 high technology goods. Evidence on this channel is, for example, provided by Caselli and Coleman (2001). They show that high levels of educational attainment are important determinants of adopting computer technology. Given the positive correlation between schooling and returns to experience as well as schooling and openness, controlling for the average level and the quality of schooling in the second stage regression is likely to reduce the coe˘ cient on openness. Information on average years of schooling in the home country is taken from Barro and Lee (1993). This variable is only available for a subset of the sample. Regressing log Open on estimated returns to home country experience in the sample for which data on average years of schooling is available does not substantially alter the coe˘ cient on log Open, as can be seen in speci...cation (I) of Table 6. As expected, returns to experience are higher in countries that have a higher level of average schooling. Controlling additionally for average years of schooling reduces the coe˘ cient on log Open. The coe˘ cient on aver- age years of schooling is, however, only signi...cant when using the entire sample and as long as GDP per capita is not added to the set of explana- tory variables. Similarly, other measures of the quantity of schooling, e.g. the log of average years of schooling and the percentage of the population with primary and secondary school completed, are not signi...cantly associ- ated with estimated returns to experience when controlling for openness and GDP per capita. Average years of schooling is the variable that maximizes the R-square. Measures of school quality, such as the quality indices from Hanushek and Kim (1999) and measures such as the pupil-teacher ratio and real expenditure per pupil from Barro and Lee (1993) do not yield statis- tically signi...cant results. Since the coe˘ cients on measures of quality and quantity of schooling turn insigni...cant once GDP per capita and openness are controlled for and since adding them to the regression has only a negli- gible e¤ect on the R-square, the following results will be presented without controlling for measures of schooling.21 Not only GDP per capita and educational attainment are likely to a¤ect the return to home country experience. On-the-job accumulation of human capital occurring in the manufacturing sector is likely to be more highly valued by the US labor market than skills acquired in the agricultural sector. Conditional on regional dummies, the partial correlation coe˘ cient between 21The R-square of a regression of returns to experience on log Open, log of GDP per Capita and regional dummies in the sample that provides information on average years of schooling is 0.303 for all countries, 0.295 for the non-Oil countries and 0.133 for the non-OECD/non-Oil Countries. 14 the share of manufacturing and returns to home country experience equals 0.25 and is signi...cant at the 5 percent level. The share of agriculture is - as expected- negatively correlated with these returns. Estimation results presented in Table 7 show that independent of the sample the coe˘ cient on the share of manufacturing is positive. Apart from non-OECD sample, this e¤ect is signi...cant, when controlling for GDP and regional dummies, indicating that on average returns to home country ex- perience are higher for immigrants proceeding from countries with a higher share of manufacturing. The e¤ect of an increase in the share of manufac- turing is small compared to the e¤ect of an increase in openness. Raising the share of manufacturing by 0.1, which is equal to one standard devia- tion, increases the predicted return to home country experience by 0.005 percentage points. An increase in Open from 0.2 to 0.3, on the other hand, raises these returns by 0.41 percentage points. The marginal e¤ect of Open on returns to experience declines as openness increases. Still, raising the trade to GDP share from 0.8 to 0.9 implies that the return to home country experience is predicted to be 0.11 percentage points higher. Amounting to 0.24 the standard deviation of openness exceeds substantially the standard deviation of the share of manufacturing. Increasing openness from 0.2 by one standard deviation raises predicted returns to home country experience by 0.8 percentage points. In none of the speci...cations presented in Table 7 the coe˘ cient on the share of agriculture is signi...cant. Institutions and government may provide an environment to individuals and ...rms that encourages the accumulation of skills and the investment in new technologies (Hall and Jones (1999)). To capture this notion of governmental and institutional quality, I follow Hall and Jones (1999) in using an index of government antidiversion policy (GADP). The index is described in detail in section A:2:2: of the Data Appendix. It assumes values between zero to one and increases in value with the e¤ectiveness of government policies in supporting an environment favorable to productive activities. Among the countries with the highest GADP ...gure Switzerland (GADP of 1), Netherlands (0.988), Sweden (0.987) and New Zealand (0.986). GADP is lowest for Liberia (0.197), Iraq (0.226) and Haiti (0.236). The results of speci...cation (IV ) in Table 7 con...rm a positive e¤ect of governmental quality on returns to experience. The e¤ect is sizeable. An increase of GADP by one standard deviation (0.2) raises returns by 1.6 percentage points in the entire sample and by 2.1 percentage points for US immigrants proceeding from non-OECD countries. The strong e¤ect of GADP on returns to experience is in line with the results of Alcala and Ciccone (2004). They ...nd that institutional quality is a highly signi...cant 15 determinant of human capital. Adding GADP to the set of explanatory variable leads to a drop in the coe˘ cient on log Open. The coe˘ cient of 0.512 in the sub-sample of non-OECD/non-Oil countries (speci...cation IV ) indicates that an increase in Open from 0.2 to 0.3 raises returns to experience of US immigrants by 0.2 percentage points. When additionally controlling for the share of manufacturing, the coe˘ cients on GDP per capita and the share of manufacturing become insigni...cant. The same holds for all regional dummies. The coe˘ cient on openness, however, remains marginally signi...cant for non-Oil countries and non-OECD/non-Oil countries. High investment rates are a precondition for technology adoption and hence learning by doing.22 It is henceforth not surprising that investment rates per worker have a signi...cantly positive e¤ect on returns to experience of US immigrants as shown in Table 8. Adding investment rates per worker to the regression, renders the coe˘ cient on GDP per capita insigni...cant and negative as can be seen in speci...cations (III) and (IV ). This is likely to be the result of a strong collinearity between GDP per capita and in- vestment per worker. The correlation coe˘ cient between GDP per capita and investment per worker amounts to 0.925 for both the entire sample and the sample of non-oil exporting countries. Regressing log of investment per worker on log of Open, log of GDP per capita and regional dummies leads to an R-square of 0.88. Only 12 percent of the variation in log investment per worker is independent of the included explanatory variables. The positive correlation of both GDP per capita and investment per workers with returns to experience combined with the high correlation among the two variables may explain the negative sign of the partial regression coe˘ cient of GDP per capita in this speci...cation. Since investment per worker is likely to be related to institutional and governmental quality, GADP is added to the regressions in the last two columns of Table 8. Log investment per workers turns insigni...cant once GADP is controlled for, as can be seen by comparing column (III) with (V ). These results indicate a positive and signi...cant e¤ect of openness on returns to home country experience of US immigrants, even when restricting the sample to immigrants from non-OECD countries. This is consistent with the hypothesis that trade increases human capital accumulation in less developed countries through technology transfer. Technology transfer may a¤ect human capital accumulation if implementing and working with new 22See literature on embodied technological progress, for example, Greenwood et al. (1997). 16 technologies increases the knowledge of workers. The hypothesis that openness leads to on-the-job learning by doing through technology transfer, may be tested by relating measures of technol- ogy transfer to estimated return to experience. Technology transfer takes place either through the production of goods in less developed countries which were already produced in more developed economies, through the importation of intermediate goods or R&D spillover. R&D spillover are likely to work through imported goods from more developed countries.23 This suggests to use technologically intensive imports from more developed economies as a proxy for technology transfer. But technology transfer may not only be related to imports, but also to exports. Exporting to more devel- oped countries may require to implement strategies that increase ...rm-level e˘ ciency. In addition, contact with foreign customers is likely to create an environment of learning opportunities. If trade leads to on-the-job human capital accumulation through tech- nology transfer, then for less developed countries imports from or exports to non-OECD countries can be expected to have a positive e¤ect on returns to experience. The same applies to imports of technology intensive goods. Re- gressing the estimated returns to experience on various measures of imports and exports, such as exports and imports by trading partner (OECD ver- sus non OECD) and imports of computers per worker yields the coe˘ cients presented in Table 9:24 Two speci...cations are displayed in this table. Speci- ...cation (I) controls for a measure of openness and regional dummies. GDP, the share of manufacturing and GADP are added in speci...cation (II). The e¤ect on returns to home-country experience remains positive independent of the trade measure used. The coe˘ cient on computer imports is statisti- cally highly signi...cant in both samples in speci...cation (I). The same applies to manufacturing exports. Summarizing, these results indicate a positive and signi...cant e¤ect of openness in the home country on returns to home country experience of US immigrants. High investment rates are a signi...cant determinant of on-the- job human capital accumulation as well as governmental and institutional 23For example, Coe, Helpman and Ho¤maister (1995) ...nd that total factor productivity in developing countries is positively associated with R&D expenditure abroad and that the spillover from an industrial country to a developing country are proportional to the share of the industrial country's imports in the developing countries'gross domestic product. 24Trade measures are taken from Caselli and Coleman (2001) who construct the data using information provided by Feenstra, Lipsey and Bowen (1997). For a detailed de- scription of the data see Caselli and Coleman (2001). With the exception of log Open all variables are de...ned in per worker terms. 17 quality. The e¤ect of trade on on-the-job human capital accumulation re- mains positive when restricting the sample to immigrants from non-OECD countries. This ...nding supports those theories that claim that opening up to trade leads to technology transfer, thereby creating learning opportunities in less developed countries. 6 Robustness Checks 6.1 Self-Selection The result that US immigrants proceeding from more open countries have a higher return to home country experience, does not necessarily allow to conclude that trade increases on-the-job human capital accumulation of the average home country resident. US immigrants do not form a random sample of the home country population. The decision to migrate is, among other things, determined by a comparison of earnings opportunities across home and destination country. As a consequence, the same covariates that a¤ect earnings, such as home country experience, schooling and ability do also a¤ect the probability to migrate. The US return to home country experience of US immigrants does therefore not necessarily have any predictive value about the US return to home country experience of the average home country resident. A short sketch of the Roy Model along the lines of Borjas (1987, 1998) illustrates this nicely. Suppose that earnings of residents in country j are given by w0j: If the entire home country population were to move to the United States, their earnings distribution in the US would be given by w1j. w0j and w1j may be written as ln w0j = + 0j 0jXj + v0j ln w1j = + 1j 1jXj + v1j For the simplicity of the exposition, let's refer to Xj as home country experience. Assume that the return to home country experience does not vary with years of experience. 0j is the return to home country experience in the home country, while 1j is the US return to home country experience of the average home country resident. Residents of source country j decide to migrate, i.e. I > 0, if wages in the US net of migration costs cj exceed wages in the home country.25 This 25Not only the di¤erences between US and source country earnings opportunities as 18 migration condition may be written as Pr ob(I > 0) = Pr ob(ln w1j cj ln w0j > 0) = Pr ob( + ( 1j 0j 1j 0j)Xj cj + v1j v0j > 0) Conditional on a given level of home country experience, the probability to migrate increases the higher the US return to home country experience relative to the home country return, holding all other wage determinants constant. Under the assumption that v0j and v1j have a bivariate normal distribution with zero mean, standard deviations 0 and 1 and correla- tion coe˘ cient , the substitution of the migration condition into the wage equation yields 0j 1j 1j E(ln w1jjXj; Ij > 0) = + ( ) 1j 1jXj + j j vj 0j where j = E(vjjvj > zj) = (zj)=(1 (zq, vj =0j(v1j: j) 1j v0j)= vj, zj=( ( and = 2 is called 0j 1j + cj 1j 0j)Xj)= vj vj vj j the inverse Mills ratio. 1 (zj) is the probability to migrate. (Heckman (1979)) The fact that the truncation of the error term depends on Xj implies that the expected value of the estimated return to home country experience of US immigrants di¤ers from the US return to home country experience of the average home country resident if j is not controlled for, as 1j + j Cov( ; Xj)=V ar( ) j j (7) where 0j 1j j j = ( ): j vj E( bj)10j= Equation (7) shows that the expected value of the return to home country of two terms. There is a direct e¤ect of home country experience on log experience estimated using the sample of US immigrants E( bj) is composed well as migration costs determine the decision to migrate. US immigration laws matter as well. Furthermore, the decision to migrate to the US is not exclusively based on economic gains, but also on family ties and political reasons. Census data does not allow to control for these factors as it neither provides information about the legal status of the immigrant nor the reason for immigration. The Roy model further assumes that migration decisions are irreversible and hence ignores the issue of self-selection induced by selective return migration. Last, self-selection may also occur when immigrants decide in which US division to reside. The regional variation in demand for skills, however, seems to be a less important determinant for the settlement of immigrants across regions than it is for native borns (see Bratsberg and Terrell (2000)). 19 earnings which is given by 1j, the parameter of interest. The second term captures the fact that home country experience a¤ects the probability to migrate. Not controlling for this term implies that the expected value of returns to home country experience estimated on the sample of US immi- grants does not allow to identify the US return to home country experience of the average home country resident. What is the direction of this bias and more importantly how does it relate to the estimated e¤ect of openness on returns to experience?26 Mi- gration theory proposes and the empirical literature provides evidence that immigrants are positively self-selected. Given positive self-selection, E(lnw1jjXj; Ij > 0) > E(lnw1jjXj) and hence j > 0. According to the Roy model the sign of the covariance between the inverse Mills ratio and home country experience Cov( ;Xj) depends on j whether the US return to home country experience exceeds or falls short of the home country return to home country experience. If the US return to home country experience exceeds the home country return, then the gain from migrating increases with home country experience and immigrants with more home country experience are likely to be drawn from a wider distribu- tion of unobservable skills. Therefore, the truncation point and j decreases as Xj rises and given positive selection, the return to home country experi- ence estimated on a sample of US immigrants is lower then 1j : But this is not the whole story. Experience of US immigrants is composed of home country experience and experience obtained when living in the US. The return to labor market experience varies according to whether the experience was acquired before or after migrating to the US. While in my sample the estimated return to home country experience evaluated at ten years of experience is 1.4 percent, the return to time spent in the US is 2.1 percent. It is a well documented fact that the return to home country experience is lower than the return to US experience of US immigrants (e.g. Chiswick and Miller (2000) and Borjas (1998)).27 26Throughout this section I refer to the term bias as the di¤erence between the expected value of returns to home country experience estimated on the sample of US immigrants and the US return to home country experience in the home country population. 27The relatively low return to pre-migration experience leads to an earnings disadvan- tage for immigrants relative to natives with similar labor market experience at arrival to the US. However, the fact the post-migration returns to experience tend to exceed returns to experience of US natives implies that with time spent in the US the relative earnings position of immigrants improves. 20 Since the decision to migrate is a function of the expected discounted life-time gain from migration, the di¤erence between pre and post-migration returns to experience determines the age at immigration. The fact that the return to post-migration experience exceeds the return to pre-migration ex- perience may lead to a decrease in the probability to immigrate to the US as home country experience increases even if US returns to home country experience exceed home country returns. This would induce a positive corre- lation between j and Xj and an upward bias of the return to home country experience estimated on the subsample of US immigrants. US immigrants from a given country are on average younger than poten- tial immigrants de...ned as home country residents who are between 15 and 64 years old. When arriving to the US, immigrants are on average 30 years old which compares to a mean age of 34 in the potential immigrant popula- tion.28 US immigrants from all countries in the sample, with the exception of South Korea and South Africa, are younger at arrival then the potential immigrant population in the home country. Cross-country variation in mean age at arrival for immigrants is substantial, ranging from 26 year for Mexico and Saudi Arabia to 34 year for South Korea. The probability to migrate for immigrants who completed their educa- tion in the home country is hump-shaped with respect to age. Constructing migration rates conditional on age and male gender for eleven ...ve-year age brackets reveals that for 67 out of 81 countries in the sample on migration rates the probability to migrate is highest for immigrants who are between 25 and 29 years old.29 Immigrants proceeding from Bahamas, Costa Rica, El Salvador, Guatemala, Mexico, Puerto Rico, Yemen, Saudi Arabia, Ire- land, Sierra Leone and Saudi Arabia are most likely to immigrate between the age of 20 and 24. For immigrants from Sri Lanka, Bulgaria and Poland the probability to migrate is highest when they are between 30 and 34 years old. The cross-country di¤erence in the age that maximizes the probability to migrate may reect di¤erences in the level of education across immigrant groups. The average Bulgarian and Sri Lankan immigrant, for example, is far more educated then the average US immigrant, while immigrants from El Salvador, Guatemala, Mexico, Puerto Rico and Yemen fall within the 28Data on population within given age brackets is taken from the International Data Base of the US Census. 29These migration rates are constructed by estimating the number of immigrants at di¤erent age brackets living in the United States based on data of the 1980 and 1990 Census. By combining these estimates with data on the population within the same age bracket. Data on the population within age brackets is taken from World Development Indicators of the World Bank. 21 group with the lowest average educational attainment. Controlling for the bias of the estimated return to home country experi- ence induced by self-selection in the ...rst-stage regression requires to account for the truncation of the error term. Adding a selection correction term in the form of the inverse Mills ratio to the Mincerian earnings equation in the ...rst step may solve this issue if the unobservables have a joint normal distribution. This Mills ratio can be constructed by using the estimated migration rates conditional on age and male gender. Since education de- termines potential experience as well as age at arrival, I also use migration rates of male immigrants for three levels of schooling: less then seven years, seven to twelve years and more than twelve. These migration rates have been taken from Bratsberg and Terrell (2002).30 Controlling for selection does not change the e¤ect of openness on re- turns to experience when controlling for log Open, log GDP per capita and signi...cant regional dummies. This can be seen by comparing speci...cation (I) of Table 10 with speci...cation (IV ) of Table 5. Once GADP and the share of manufacture are added to the regression in speci...cation (II) the coe˘ cient on log Open decreases by about 10 percent in the entire sam- ple and by 30 percent in the non-OECD sample relative to the uncorrected coe˘ cients if migration rates conditional on age are used to control for self- selection. Migration rates conditional on schooling are only available for a subset of countries. Adding them to the ...rst step regressions reduces the sample size signi...cantly as can be seen in the last two columns of Table 10: Despite the change in the sample, the coe˘ cients on log Open are similar in speci...cation (I) independent on whether migrations rates conditional on age or conditional on education are added to the ...rst step regression. 6.2 Returns to Experience and Schooling Log earnings experience pro...les tend to be steeper for better schooled work- ers. An OECD study, for example, suggests that participation in job- related training programs is correlated with educational attainment (OECD 30Relaxing the assumption of joint normality in order to identify the US return to home country experience in the home country population requires to satisfy an exclusion restriction,i.e. a regressor that a¤ects the migration probability of immigrants from a given country di¤erently, but does not determine earnings. As age at arrival does not only determine the probability to migrate but also earnings, the exclusion restriction is not satis...ed. Immigrants may leave their country for political reasons. Wars or political turmoil are likely to a¤ect the decision to migrate, but not wages. If these variables a¤ect immigrants who immigrate at di¤erent years of immigration in a di¤erent way, they may be used in order to control for selection. 22 ( 1997)). Similarly, Bartel and Sicherman (1998) provide empirical evidence that the probability of receiving training increases monotonically with ed- ucation. Psacharopolous and Layard (1979) show that experience pro...les are steeper for individuals with higher educational attainment. Using US panel data, Altonji and Pierret (1997) reach a similar conclusion. Based on data of 11 European countries, Brunello and Comi (2000) ...nd that employ- ees with tertiary education have steeper experience pro...les then employees with upper secondary or compulsory education. The same applies to US immigrants. Returns to home country experience of US immigrants are higher for immigrants with more years of education. Table 11 shows that the slope of the home country experience pro...le is steeper for US immigrants with at least a high school degree as compared to immigrants with a lower level of schooling. At the same time the coe˘ cient on the square of experience is higher for better schooled immigrants indi- cating a faster decrease in returns to experience. The largest gap between skilled and less skilled US immigrants is reached after approximately ten years of labor market experience. Controlling for ...xed e¤ects decreases the slope of the log earnings pro...le slightly. It leaves the fact unchanged that the log earnings pro...le is steeper for younger workers with higher educational attainment. If returns to home country experience vary with education then the OLS estimator of the homogenous US return to home country experience may be written as a variance weighted average of the returns to home country experience of the di¤erent educational categories. As US immigrants are on average better educated then home country residents, heterogeneity in returns to experience implies that the average US return to home country experience estimated on a sample of US immigrants exceeds the average US return to home country experience in the home country population.31 If education a¤ects returns to experience and if educational di¤erences between home country and immigrants population are related to source country characteristics, then the estimated e¤ect of openness on returns to experience will di¤er from the e¤ect of openness on US returns to home country experience of the average source country resident. Consider, for ex- 31The concavity of the log earnings-experience pro...les implies that the return to ex- perience decreases with years of experience. There exists empirical evidence that this decrease is more pronounced for workers with a higher level of schooling (see for example, Brunello and Comi (2000)). Workers with a long labor market experience and a high level of schooling may therefore have lower returns to experience then comparable less schooled workers. Consequently, the claim that returns to experience are higher for more schooled workers refers in what follows to workers with few years of labor market experience. 23 ample, that openness a¤ects the decision to migrate by reducing migration costs because it familiarizes residents in the home country with the institu- tional, cultural and social environment in the United States. The decrease in migration costs is likely to be larger for workers with a high level of school- ing relative to workers with a low level of schooling because of di¤erences in literacy, language and technological skills. Openness may then have a larger e¤ect of the migration probability of better schooled workers relative to less schooled workers. Given that returns to experience are higher for immigrants with more schooling, we would observe a positive correlation between openness and returns to experience because openness induces more skilled workers to leave the country, and not because openness induces more accumulation of on-the-job human capital. But opening up to trade may also induce a decrease in the relative mi- gration rate of skilled workers. As pointed out above, there exists empirical evidence that trade increases the demand for skilled workers. If openness raises the return to schooling or returns to home country experience in the home country relative to the respective return in the United States, open- ness may actually lead to a decrease in the immigration rate of highly skilled workers. In this case, the e¤ect of openness on returns to home country ex- perience would be underestimated.32 The positive e¤ect of openness on returns to home country experience is not determined by the skill distribution of the US immigrant population. This is demonstrated by regressing returns to home country experience of highly skilled and low skilled workers separately on log Open and other control variables in the ...rst and second column of Table 12.33 Comparing these coe˘ cients with column (V I) in Table 5 reveals that the coe˘ cient 32Di¤erence in educational attainment between US immigrants and home country pop- ulation varies substantially across countries, ranging from 2 to 18 years. It is largest for immigrants proceeding from Africa. US immigrants, for example, from Namibia and Mali form the group with the highest educational attainment of US immigrants with on av- erage 18 years of schooling. At the same time educational attainment in these countries is among the lowest in the world. On the other hand, educational attainment of Mexi- can, Canadian, Italian, Greek and Portuguese US immigrants mimics rather well average educational attainment in their home country population. 33Highly skilled workers are de...ned as workers with at least a high school degree. The average return to experience of low skilled workers in this sample amounts to 2.18 percent (with a standard deviation of 2.587), while the corresponding return of highly skilled workers equals 2.965 percent (4.893). Estimations presented in Table 12 exclude countries with less then 100 observations in order to increases the precision of the estimates which are obtained in the ...rst step. This implies that Algeria, Liberia, Malta, Saudi Arabia, Sierra Leone, Singapore, Switzerland, Tanzania, Tunisia, Uganda and Yemen drop out of the sample. 24 on log Open does not change signi...cantly in the entire sample as well as the sample of non-oil countries. But the e¤ect of openness on returns to experience of highly skilled US immigrants - as shown in column (II) - is not signi...cant in any of the three samples. Apart from GADP, no explanatory variables is signi...cant in this speci...cation arising from the fact that the cross-country variance in returns to experience is substantially higher for more skilled workers. Since the variance of the returns to experience among highly skilled is relatively large, the mean square error of speci...cation (II) is about four times higher than in speci...cation (I), scaling up the standard errors of the coe˘ cients. Using the average estimated return to home country experience evalu- ated at the skill distribution in the home country does also not alter the principal ...ndings. Column (III) of Table 12 uses average estimated returns to home country experience evaluated at the skill distribution in the home country as dependent variable. The results are very similar to the estimates presented in Table 7 for the entire sample and the sample of non-Oil coun- tries. Increasing the share of trade in GDP from 0.2 to 0.3, raises the average return to home country experience in the non-Oil economies by about 0.35 percent. However, when restricting the sample to non-OECD economies the coe˘ cient on Open is about 50 percent lower than the respective coe˘ cient of Table 7. The coe˘ cient on log Open does also not change signi...cantly when evaluating returns to experience at the skill distribution of US immi- grants as can be seen in the last column of Table 12. Summarizing, these ...ndings reveal that the e¤ect of openness on returns to experience is not driven by cross-country di¤erences in the educational attainment di¤erential between home country residents and US immigrants. Controlling for heterogeneity in returns, however, decreases the coe˘ cient on openness in the non-OECD sample and renders it insigni...cant. 6.3 English Speaking Origin Returns to home country experience depend on whether the immigrant can transfer his knowledge and skills to the US economy. Transferability of human capital to the US is largely determined by English pro...ciency. Since countries where English is an o˘ cial language tend to be more open, the estimated coe˘ cient on openness may overestimate the e¤ect of openness on returns to home country experience. The ...nding of a positive and signi...cant e¤ect of openness on returns to experience applies to immigrants from English speaking and non Eng- lish speaking countries alike. The results presented in Table 13 provide 25 evidence that the positive e¤ect of openness on returns to home country ex- perience is independent on whether the immigrant originates from a country where English is widely spoken or not. English speaking countries subsume all countries were English is an o˘ cial language or widely used in certain population groups.34 Columns (I) and (II) of table 13 display the es- timation results for non-English speaking and English speaking countries, respectively. The coe˘ cient on openness is positive and signi...cant for both speci...cations. The di¤erence in the coe˘ cients is not statistically signi...- cant, which is likely to reect the fact that English pro...ciency is already controlled for in the ...rst step regression. 6.4 Cultural Background The question remains to be answered whether the e¤ect of openness on home country experience may capture cultural di¤erences among US immigrants. Re-estimating the ...rst step Mincerian equation for the sample of immigrants who completed their education, but had no labor market experience in their home country allows to address this issue. If the e¤ect of openness on returns to home country experience is determined by cultural di¤erences, openness can be expected to exert a signi...cant and positive e¤ect on returns to US experience of US immigrants who had at least some exposure to the culture of their home country. Controlling only for openness leads to coe˘ cients close to zero inde- pendent of the sample, as can be seen in column (I) of Table 14. Adding signi...cant regional dummies in column (II) and GDP per capita in column (III) increases the coe˘ cients on log Open in the di¤erent samples. Still they remain insigni...cant. Once GADP and the share of manufacturing are added to the regression the coe˘ cient on openness falls to zero again. Cul- tural background is therefore unlikely to explain the e¤ect of openness on returns to home country experience. 7 Conclusion This paper provides empirical evidence that trade increases on-the-job hu- man capital accumulation. This ...nding is not the result of self-selection, heterogeneity in returns to experience, English speaking origin or cultural background. The e¤ect of trade on on-the-job human capital accumula- tion remains positive when restricting the sample to immigrants from non- 34The set of English speaking countries is described in the Data Appendix. 26 OECD countries, supporting the claim that trade leads to technology trans- fer, thereby creating learning opportunities in less developed countries. Human capital accumulation is considered an important determinant of economic growth. While a considerable amount of research has been dedi- cated in explaining cross-country di¤erences in the accumulation of human capital in school and research organizations, less is known about on-the-job accumulation of human capital. But on-the-job human capital accumulation is likely to contribute con- siderably to economic growth. Historically, the acquisition of human capital mainly took place on-the-job. The extended schooling system and the conse- quent late entry into the labor market are a phenomenon of the last decades. Moreover, high educational attainment is a characteristic of rich countries. For many less developed economies, the main bulk of human capital ac- cumulation is still likely to occur on-the-job. If trade increases on-the job human capital accumulation, its role in generating growth is likely to be more important than generally considered. 27 A Data Appendix A.1 Sample Selection Criteria The analysis is based on a sample drawn from the 5/100 public-use micro data ...les of the US Censuses of Population of 1980 and 1990.35 The sam- ple is restricted to male US immigrants who are between 25 and 64 years old, worked and earned at least $1000 wage or salary income in the year preceding the census and were not enrolled in school at the time of the cen- sus. Furthermore, immigrants in the sample arrived to the US after 1959 and completed their education in their home country. The latter restriction is motivated by the fact that both censuses provides only information on the highest educational degree obtained. It is hence impossible to identify years of experience in the home country for immigrants who acquired US schooling. The censuses provides information on the year of immigration only within brackets of varying width. Following Bratsberg and Ragan (2002), Brats- berg and Terrell (2002) and Chiswick and Miller (2002) an immigrant is included in the sample if (6+ years of education) is lower than his age at the lower bound of the year of immigration bracket. This restriction ensures that no immigrant who acquired US schooling enters the sample. These sample selection criteria leave a total sample of 173137 observations, of which 58695 belong to the 1980 census and 114442 to the 1990 census. Descriptive statistics of the full sample are presented in Table 1. The sample size reduces to 171445 observations when matching the cen- sus data with the Penn World Tables Mark 5.6 as some countries do not report information on openness. A.2 Variable Description A.2.1 Census Variables The dependent variable of the wage regression is the natural logarithm of the annual wage or salary income in 1979 or 1989. Years of schooling in the 1980 census are based on the "Highest Year of Schooling Attended". If the respondent did not complete the highest grade attended, one year is subtracted. The rule used to convert educational attainment to years of schooling in the 1990 census is the same as in Bratsberg and Terrell (2002). 35The data are available online at http://www.ipums.org. For more information on the data, see Ruggles and Sobek (1997). 28 Experience is de...ned as (age years of schooling 6): Experience in the US is calculated with respect to the middle of the year of immigration bracket. A.2.2 Macroeconomic Variables Variables necessary to calculate the measure of GDP per capita, investment per worker and Open are taken from the Penn World Tables Mark 5.6 revi- sion of Summers and Heston (1991). The measure of GDP is RGDPL, which is per capita GDP expressed in constant year international prices. Open is de...ned as imports plus exports in exchange rate US$ relative to GDP in purchasing-power-parity US$ and deated by international export prices. This measure of openness has been proposed by Ciccone and Alcala (2004). Average educational attainment is measured for the population aged 25 and over, as reported by Barro and Lee (2000). Data on pupil-teacher ratio and real public spending per student are from Lee and Barro (2001). Measures of cognitive skills are taken from Hanushek and Kim (1999). The index of government antidiversion (GADP) is taken from Hall and Jones (1999). It is based on data from the International Country Risk Guide which rates 130 countries according to 24 categories. The GADP is de...ned as the equal-weighted average of ...ve of these categories (law and order, bureaucratic quality, corruption, risk of expropriation and government repudiation of contracts) for the years 1986-1995. The index is measured from zero to one. The value of the index increases with the e¤ectiveness of government policies in supporting an environment favorable to productive activities. Data on the share of agriculture and manufacturing in GDP are taken from Caselli and Coleman (2001) and are based on data from the World Bank. Data on imports and exports are from Caselli and Coleman (2001), who take the original data from Feenstra, Lipsey and Bowen (1997). A.2.3 Regional Dummies Regional Dummies are de...ned for Africa, Asia, Latin America, Transition Economies and Island. Island includes the Caribbean and Paci...c Island States, the African Island States in the Indian Ocean as well as Cape Verde, Malta and Cyprus. Hongkong, Taiwan and Singapore are added to the Asian dummy. Oil exporting countries are Iran, Iraq, Jordania, Saudi Arabia, Syria and Yemen. The base dummy consists of the OECD member countries as of 1990 plus Turkey and Israel. 29 A.2.4 English Speaking Countries According to the World Factbook 2001 English is an o˘ cial language in Australia, Bahamas, Barbados, Belize, Canada, Fiji, Ghana, Guyana, Hong Kong, Ireland, Jamaica, Kenya, Liberia, Malta, New Zealand, Nigeria, Philip- pines, Singapore, South Africa, Tanzania, Trinidad and Tobago, Uganda and the UK. Countries were English is an o˘ cial language but used by a limited minority are Sierra Leone, India and Pakistan. Countries were English is not an o˘ cial language but widely used within certain population groups applies to Jordania, Panama, South Korea and Western Samoa. 30 References [1] Alcala, Francisco and Antonio Ciccone (2004) "Trade and Productiv- ity", Quarterly Journal of Economics, Vol. CXIX, No. 2, pp. 613-46. [2] Altonji, Joseph and Charles Pierret (1998) "Employer learning and the signalling value of education", in Ohashi and T. Tachibanaki (eds.) Internal Labour Markets: Incentives and Employment. McMillan Press. pp. 159-195. [3] Barro, Robert and Jongwha Lee (2000) "International Data on Educa- tional Attainment: Updates and Implications", February, manuscript, Harvard University. [4] Barro, Robert and Jongwha Lee (1993) "International Comparisons of Educational Attainment", Journal of Monetary Economics, XXXII, pp. 363-394. [5] Bartel, Ann and Nachum Sicherman (1998) "Technological Change and the Skill Acquisition of Young Workers", Journal of Labor Economics, 16, pp. 718-55. [6] Bartel, Ann and Frank Lichtenberg (1987) "The Comparative Advan- tage of Educated Workers in Implementing New Technology", Review of Economics and Statistics, vol. 69, no.1, pp. 1-11. [7] Ben-Porath, Yoram (1967) "The Production of Human Capital and the Life Cycle of Earnings", Journal of Political Economy, vol. 75, pp. 352-365. [8] Bleakley Hoyt and Aimee Chin (2002) "Language Skills and Earnings: Evidence from First-Generation Immigrants", mimeo. [9] Borjas, George (1998) "The Economic Progress of Immigrants", NBER Working Paper, No. 6506, Cambridge, MA. [10] Borjas, George (1987) "Self-Selection and the Earnings of Immigrants", American Economic Review, 77 (4), pp. 531-553. [11] Bratsberg, Bernt, and James F. Ragan (2002) "The Impact of Host- Country Schooling on Earnings - A Study of Male Immigrants in the United States", Journal of Human Resources , vol. 37, no. 1, Winter, pp. 63-105. 31 [12] Bratsberg, Bernt, and Dek Terrell (2002) "School Quality and Returns to Education of US. Immigrants", Economic Inquiry, vol. 40, no. 2, pp. 177-198. [13] Brunello Giorgio and Simona Comi (2000) "Education and Earnings Growth: Evidence from 11 European Countries", Fondazione ENI En- rico Mattei Working Paper, no. 00-29. [14] Card, David and Alan Krueger (1992)"Does School Quality Matter? Returns to Education and the Characteristics of Public Schools in the United States", Journal of Political Economy, vol. 100, no. 1, pp. 1-40. [15] Caselli, Francesco and Wilbur John Coleman (2001) "Cross-Country Technology Di¤usion: The Case of Computers", American Economic Review, Papers and Proceedings, May, vol. 91, no. 2, pp. 328-335. [16] Chiswick, Barry and Paul Miller (2002) "The Complementarity of Lan- guage and Other Human Capital: Immigrant Earnings in Canada", IZA Discussion Paper, No 451, Institute for the Study of Labor (IZA), Bonn. [17] Chiswick, Barry and Paul Miller (1995) "The Endogeneity between Language and Earnings: International Analyses", Journal of Labor Eco- nomics, vol. 13, pp. 246-288. [18] Chiswick, Barry (1978) "The E¤ect of Americanization on the Earnings of Foreign-born Men", Journal of Political Economy, October, vol. 86, no. 5, pp. 897-921. [19] CIA (2001) "The World Factbook 2001", Washington , D.C. [20] Chuang, Yih-chyi (2000) "Human Capital, Exports and Economic Growth: A Causality Analysis for Taiwan, 1952-1955", Review of In- ternational Economics, vol. 8, no. 4, pp. 712-720. [21] Coe, D., E. Helpman and A.W. Ho¤maister (1997) "North-South R&D Spillovers", The Economic Journal, vol. 107, pp. 134-149. [22] Denny, Kevin, Colm Harmon and Reamonn Lydon (2001) "Cross- Country Evidence on the Returns to Education: Patterns and Expla- nations", mimeo. [23] Dollar, David and Aart Kraay (2002) "Institutions, Trade and Growth", The World Bank, mimeo, Washington, D.C. 32 [24] Feenstra, Robert, Robert Lipsey and Harry Bowen (1997)"World Trade Flows, 1970-1992. With Production and Tari¤ Data", NBER Working Paper, no. 5910, Cambridge, MA. [25] Frankel, Je¤rey and David Romer (1999) "Does Trade Cause Growth?", American Economic Review, June, vol. 89, no. 3, pp. 379-399. [26] Goh Ai-Ting Goh and Jacques Olivier. (2002) "Learning by Doing, Trade in Capital Goods and Growth", Journal of International Eco- nomics, vol. 56, pp. 411-444. [27] Greenwood, Jeremy, Zvi Hercovitz and Per Krusell "Long-run impli- cations of Investment-Speci...c Technological Change", American Eco- nomic Review, June 1997, vol. 87, no. 3, pp. 342-62. [28] Hall, Robert and Charles Jones (1999) "Why Do Some Countries Pro- duce Do Much More Output Per Worker Than Others?", Quarterly Journal of Economics, February, vol.114, pp. 83-116. [29] Hanushek, Eric and Jin-Yeong Kim (1999) "Immigration and Home- Country School Quality", November 1999, mimeo. [30] Hanson, Gordon and Ann Harrison (1999) "Trade, Technology, and Wage Inequality in Mexico", Industrial and Labor Relations Review, vol. 52, pp. 271-288. [31] Heckman, James (1979) "Sample Selection Bias as a Speci...cation Er- ror", Econometrica, January, vol. 47, no. 1, pp. 153-161. [32] Jasso, Guillermina and Mark Rosenzweig (1990) "Self-Selection and the Earnings of Immigrants: Comment", American Economic Review, vol. 80, no. 1, pp. 298-1990. [33] Jovanovic, Boyan (1979) "Firm-Speci...c Capital and Turnover", Journal of Political Economy, vol. 87, pp. 1246-1260. [34] Lazear, Edward (1981) "Agency, Earnings Pro...les, Productivity and Hours Restrictions", American Economic Review vol. 71, pp. 606-620. [35] Lee, Jong-Wha and Robert J. Barro (2001) "Schooling Quality in a Cross-Section of Countries", Economica, vol. 68, pp. 465-488. [36] Lucas, Robert (1993) "Making a Miracle", Econometrica, vol. 61, no. 2, pp. 251-272. 33 [37] Lucas, Robert (1988) "On the Mechanics of Economic Development", Journal of Monetary Economics, vol.22, pp.3-42. [38] Mincer, Jacob (1997) "The Production of Human Capital and the Life Cycle of Earnings: Variations on a Theme", Journal of Labor Eco- nomics, vol. 15, no. 1, S26-S39. [39] Moulton, Brent (1990) "An illustration of a pitfall in estimating the ef- fects of aggregate variables on micro unit", Review of Economic Studies, vol. 72, no. 2, pp. 334-338. [40] Moulton, Brent (1986) "Random Group E¤ects and the Precision of Regression Estimates", Journal of Econometrics, vol. 32, pp. 385-397. [41] Murphy, Kevin and Finis, Welch (1990) "Empirical Age-Earnings Pro- ...les", Journal of Labor Economics, April, vol. 8, pp. 202-229. [42] OECD (1997), Education Policy Analysis, Paris. [43] Pissarides, Christopher (1997) "Learning by Trading and the Returns to Human Capital in Developing Countries", World Bank Review ,vol. 11, pp. 17-32. [44] Psacharapoulous, George (1994) "Returns to Investment in Education: A Global Update", World Development, 22, pp. 1325-1343. [45] Psacharopoulos, George and Richard Layard (1979) "Human Capital and Earnings: British Evidence and Critique", Review of Economic Studies,vol. 46,pp. 485 - 503. [46] Robbins, Donald J. (1995) "Trade, Trade Liberalization, and Inequal- ity in Latin America and East Asia: Synthesis of Seven Country Stud- ies", Institute for Inernational Development, Harvard University, Cam- bridge, Mass. [47] Robbins, Donald J. (1994) "Earnings Dispersion in Chile after Trade Liberalization", Institute for International Development, Harvard Uni- versity, Cambridge. Mass. [48] Ruggles, Steven, Matthew Sobek et. al. (1997) "Integrated Public Use Microdata Series: Version 2.0 Minneapolis: Historical Census Projects", University of Minnesota. 34 [49] Salop Joanne and Steven Salop (1976) "Self-Selection and Turnover in the Labor Market", Quarterly Journal of Economics, vol. 90, pp. 619-627. [50] Stokey, Nancy (1991) "The Volume and Composition of Trade between Rich and Poor Countries", Review of Economic Studies, vol. 58, pp. 63-80. [51] Summers, Robert and Alan Heston (1991) "The Penn World Table (Mark 5): An Expanded Set of International Comparisons: 1950-1988", Quarterly Journal of Economics, CVI, pp. 327-368. [52] Young, Aylwin (1991) "Learning by Doing and Dynamic E¤ects of In- ternational Trade", Quarterly Journal of Economics, vol. 106, pp. 369- 406. 35 Tables Table 1 Descriptive Sample Statistics 1980 Census 1990 Census Variable Mean Std. Dev. Mean Std. Dev. Log of Annual Earnings 9.346 0.754 9.757 0.852 Experience 23.736 10.377 24.656 10.678 Experience US 8.884 5.220 11.493 7.406 Education Years of Education 10.108 5.180 10.022 5.362 Grade less than 5 0.150 0.357 0.174 0.379 Grade 5 to 8 0.258 0.438 0.194 0.395 Grade 9 0.040 0.196 0.044 0.206 Grade 10 to 11 0.070 0.256 0.100 0.300 Grade 12 and GED 0.195 0.396 0.153 0.360 Some College 0.024 0.154 0.096 0.294 Associate Degree 0.073 0.260 0.042 0.200 Bachelor's Degree 0.109 0.312 0.106 0.308 Master's Degree 0.025 0.156 0.048 0.214 Professional/Doctoral 0.055 0.227 0.043 0.202 Region Pacific 0.348 0.476 0.384 0.486 Mid Atlantic 0.260 0.438 0.213 0.410 East North Central 0.115 0.319 0.078 0.268 West North Central 0.013 0.114 0.010 0.101 South Atlantic 0.079 0.269 0.118 0.323 East South Central 0.005 0.072 0.005 0.071 West South Central 0.084 0.278 0.101 0.301 Mountain 0.031 0.173 0.036 0.187 New England 0.065 0.247 0.054 0.226 Year of Immigration 1960-64 0.142 0.349 0.052 0.222 1965-69 0.224 0.417 0.090 0.286 1970-74 0.304 0.460 0.141 0.348 1975-80 0.330 0.470 0.185 0.389 1980-81 0.000 0.000 0.137 0.343 1982-84 0.000 0.000 0.126 0.332 1985-86 0.000 0.000 0.120 0.325 1987-90 0.000 0.000 0.149 0.356 Others Married (Spouse Present) 0.776 0.417 0.685 0.464 English (Only or Very well) 0.379 0.485 0.396 0.489 Disability 0.023 0.148 0.025 0.155 SMSA 0.913 0.282 0.910 0.286 # Observations 58695 114442 Base dummies are Grade 12 and GED for education, Pacific for region and 1960-64 for year of immigration. For variable description, see Data Appendix. 36 Table 2 Earnings Regression Census 1980 1990 1980/1990 Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Grade less than 5 -0.288*** 0.051 -0.351*** 0.033 -0.351*** 0.033 Grade 5 to 8 -0.172*** 0.033 -0.270*** 0.02 -0.270*** 0.02 Grade 9 -0.147*** 0.021 -0.211*** 0.023 -0.212*** 0.023 Grade 10 to 11 -0.059*** 0.016 -0.134*** 0.014 -0.134*** 0.014 Some College 0.091*** 0.025 0.129*** 0.017 0.129*** 0.017 Associate Degree 0.156*** 0.023 0.240*** 0.026 0.241*** 0.026 Bachelor's Degree 0.375*** 0.059 0.444*** 0.074 0.444*** 0.074 Master's Degree 0.471*** 0.04 0.624*** 0.042 0.624*** 0.041 Profess./Doctoral 0.617*** 0.025 0.777*** 0.027 0.777*** 0.027 Exp 0.024*** 0.004 0.022*** 0.004 0.023*** 0.004 Exp2 / 100 -0.046*** 0.007 -0.041*** 0.007 -0.043*** 0.007 ExpUS 0.041*** 0.009 0.036*** 0.009 0.036*** 0.008 ExpUS2/100 -0.102*** 0.027 -0.073*** 0.016 -0.075*** 0.014 Exp*ExpUS 0.018 0.013 0.017 0.013 0.018 0.013 Married 0.252*** 0.013 0.272*** 0.016 0.272*** 0.015 English 0.203*** 0.046 0.199** 0.044 0.199*** 0.044 Disabled -0.245*** 0.033 -0.277** 0.031 -0.277*** 0.031 SMSA 0.045 0.036 0.077** 0.019 0.077*** 0.019 Mid Atlantic 0.002 0.036 0.129*** 0.039 0.129*** 0.039 East North Central 0.192*** 0.022 0.113*** 0.024 0.113*** 0.024 West North Central 0.134*** 0.046 -0.010 0.035 -0.010 0.034 South Atlantic 0.001 0.024 -0.021 0.033 -0.022 0.032 East South Central 0.140** 0.059 0.076 0.057 0.076 0.057 West South Central -0.007 0.021 -0.173*** 0.017 -0.173*** 0.017 Mountain -0.004 0.031 -0.114*** 0.016 -0.114*** 0.016 New England 0.118*** 0.037 0.224*** 0.05 0.224*** 0.05 1980 -0.447 0.092 Constant 8.435*** 0.052 8.835*** 0.067 8.826*** 0.059 Adjusted R-Square 0.253 0.313 0.333 # Observations 58695 114442 173137 Dependent variable is the log of annual earnings. Regression using 1980 and 1990 Censuses includes interaction terms between the 1980 dummy and all other dummies. Standard errors account for heteroscedasticity and clustering within country of origin. Double asterisk denotes statistical significance at the 5-percent level and triple at the 1-percent level. 37 Table 3 Earnings Regression and Openness OLS Fixed Effect Constrained Exp 0.034*** 0.025*** 0.024*** (0.004) (0.001) (0.001) Exp2 / 100 -0.039*** -0.039*** -0.027*** (0.007) (0.002) (0.001) ExpUS 0.028*** 0.027*** 0.02*** (0.008) (0.001) (0.001) ExpUS2/100 -0.059*** -0.046**** -0.100*** (0.012) (0.005) (0.005) Exp*ExpUS/100 0.005 -0.014*** 0.053*** (0.011) (0.004) (0.003) Exp*log RealOpen 0.006*** 0.002*** 0.006*** (0.001) (0,000) (0,000) ExpUS*log RealOpen -0.004*** -0.006*** -0.006*** (0.001) (0,000) (0,000) Constant 8.833*** 8.814*** 8.975*** (0.057) (0.014) (0.012) R-Square 0.344 0.313 # Observations 171445 171445 171445 Dependent variable is the log of annual earnings. Regressions include dummies for year of education, marital status English speaking status, health status, SMSA, census division, 1980 and year of immigration, as well as interaction terms between the 1980 dummy and all other dummies. Standard errors are in parentheses. They account for heteroscedasticity and clustering within country of origin. Single asterisk denotes statistical significance at the 10-percent, double at the 5- percent, triple at the 1-percent level. 38 Table 4 Estimated Returns to Home Country Experience Country Exp. Std. Err. Exp.^2 Std. Err. # Obs. Tunisia 0.173 0.255 -0.00433 0.00738 51 Singapore 0.101 0.069 -0.00213 0.00153 67 Norway 0.09*** 0.026 -0.0012* 0.00067 198 Finland 0.089*** 0.034 -0.00142** 0.00070 134 Japan 0.086*** 0.007 -0.00132*** 0.00021 2884 South Africa 0.085*** 0.019 -0.0018*** 0.00048 419 Tanzania 0.085 0.058 -0.00102 0.00134 78 Uganda 0.082 0.071 -0.00406 0.00245 80 Germany. West 0.082*** 0.010 -0.00164*** 0.00027 1750 Cyprus 0.082** 0.032 -0.00121* 0.00066 105 Malaysia 0.076*** 0.025 -0.00112** 0.00050 205 Netherlands 0.075*** 0.015 -0.0011*** 0.00038 661 Denmark 0.074*** 0.028 -0.00144** 0.00070 232 Saudi Arabia 0.068 0.056 0.00016 0.00132 84 Switzerland 0.067*** 0.023 -0.00165*** 0.00060 387 Canada 0.067*** 0.006 -0.00121*** 0.00013 4588 Belgium 0.066** 0.029 -0.00182** 0.00078 225 United Kingdom 0.066*** 0.005 -0.00122*** 0.00012 5866 Western Samoa 0.066* 0.035 -0.00146* 0.00079 141 Sweden 0.065*** 0.023 -0.00113* 0.00068 300 Australia 0.062*** 0.022 -0.00106** 0.00054 426 Ethiopia 0.052** 0.024 -0.00114* 0.00064 282 Sri Lanka 0.048 0.032 -0.00127* 0.00076 184 Israel 0.047*** 0.014 -0.00077** 0.00032 953 Brazil 0.046*** 0.012 -0.00066** 0.00030 782 Bulgaria 0.043 0.046 -0.00028 0.00099 130 Hungary 0.042** 0.019 -0.00046 0.00044 634 Malta 0.042 0.047 0.00005 0.00096 83 Ireland 0.041*** 0.012 -0.00084*** 0.00029 1164 Kenya 0.041 0.055 -0.00015 0.00147 129 France 0.039*** 0.015 -0.00039 0.00044 864 Czechoslovakia 0.039** 0.017 -0.00075* 0.00039 608 Indonesia 0.039** 0.017 -0.0008* 0.00040 473 Egypt 0.039*** 0.012 -0.00081*** 0.00031 1178 Algeria 0.037 0.103 0.0009 0.00280 66 Italy 0.037*** 0.005 -0.00061*** 0.00009 5373 Nigeria 0.035 0.027 -0.00007 0.00080 385 Exp. is the estimated coefficient on experience and Exp2. on experience squared. Std. Err. refers to Standard Errors. The dependent variable is the log of annual earnings. The specification of the regression is described in the text. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple at the 1-percent level. 39 Table 4 Estimated Returns to Home Country Experience continued Country Exp. Std. Err. Exp.^2 Std. Err. # Obs. Czechoslovakia 0.039** 0.017 -0.00075* 0.00039 608 Indonesia 0.039** 0.017 -0.0008* 0.00040 473 Egypt 0.039*** 0.012 -0.00081*** 0.00031 1178 Algeria 0.037 0.103 0.0009 0.00280 66 Italy 0.037*** 0.005 -0.00061*** 0.00009 5373 Nigeria 0.035 0.027 -0.00007 0.00080 385 Morocco 0.034 0.038 -0.00007 0.00101 207 Turkey 0.034*** 0.012 -0.00065**** 0.00025 575 Taiwan 0.034*** 0.009 -0.00046** 0.00021 1728 Iran 0.032*** 0.010 -0.0005** 0.00021 1644 Bolivia 0.032* 0.019 -0.00066 0.00044 298 Argentina 0.031*** 0.010 -0.00048** 0.00022 1383 Thailand 0.030* 0.016 -0.00088** 0.00040 609 New Zealand 0.027 0.032 -0.00037 0.00085 216 Hong Kong 0.027** 0.012 -0.00069** 0.00028 742 Greece 0.027*** 0.007 -0.00045*** 0.00014 2695 Korea. Rep. 0.027* 0.016 -0.00054 0.00035 640 Dominican Rep. 0.022*** 0.005 -0.00041*** 0.00010 3150 Bahamas 0.021 0.040 0.00005 0.00090 108 Portugal 0.021*** 0.004 -0.00046*** 0.00008 3985 Peru 0.021*** 0.008 -0.00047*** 0.00017 1661 Chile 0.021 0.013 -0.00038 0.00030 769 India 0.020*** 0.005 -0.0005*** 0.00011 5871 Belize 0.019 0.022 -0.00036 0.00040 241 Ecuador 0.018** 0.007 -0.0003** 0.00015 1750 Colombia 0.018*** 0.005 -0.00035*** 0.00011 3258 Barbados 0.018 0.014 -0.00058** 0.00029 509 Haiti 0.017*** 0.006 -0.00028** 0.00012 2484 China 0.017*** 0.004 -0.00031*** 0.00007 5960 Yugoslavia 0.017** 0.008 -0.00026 0.00016 1873 Puerto Rico 0.016*** 0.005 -0.00026** 0.00012 3877 Exp. is the estimated coefficient on experience and Exp2. on experience squared. Std. Err. refers to Standard Errors. The dependent variable is the log of annual earnings. The specification of the regression is described in the text. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple at the 1-percent level. 40 Table 4 Estimated Returns to Home Country Experience continued Country Exp. Std. Err. Exp.^2 Std. Err. # Obs. Mexico 0.016*** 0.001 -0.00028*** 0.00003 54610 Guatemala 0.015** 0.007 -0.00012 0.00013 2313 Poland 0.014** 0.006 -0.00026** 0.00012 3322 Romania 0.014 0.011 -0.00042* 0.00024 938 Costa Rica 0.013 0.015 -0.00011 0.00032 422 Jamaica 0.012** 0.006 -0.00022** 0.00011 3355 Syria 0.011 0.017 -0.00028 0.00032 429 Philippines 0.010*** 0.003 -0.00036*** 0.00006 10085 Uruguay 0.010 0.019 -0.00021 0.00040 381 Ghana 0.009 0.035 -0.00088 0.00087 252 Guyana 0.008 0.010 -0.00015 0.00019 1075 Jordan 0.006 0.024 -0.00016 0.00053 300 El Salvador 0.006 0.005 -0.00013 0.00009 4419 Austria 0.006 0.028 0.00037 0.00082 309 Trinidad & Tobago 0.005 0.010 -0.00001 0.00022 1183 Pakistan 0.004 0.011 0.00013 0.00028 1150 Spain 0.004 0.011 -0.00032 0.00024 986 Cape Verde Islands 0.004 0.022 -0.00007 0.00038 173 Iraq 0.003 0.015 0.00042 0.00031 553 U.S.S.R. 0.003 0.008 0.0001 0.00017 2197 Honduras 0.003 0.012 0.00000 0.00026 861 Panama 0.002 0.015 -0.00017 0.00034 575 Myanmar 0.001 0.022 0.00008 0.00045 284 Nicaragua -0.003 0.008 0.00006 0.00017 1272 Bangladesh -0.004 0.023 0.00033 0.00051 289 Fiji -0.004 0.020 0.00002 0.00038 239 Venezuela -0.007 0.027 0.00036 0.00063 273 Paraguay -0.032 0.062 0.00038 0.00160 75 Yemen -0.037 0.045 -0.00004 0.00079 86 Liberia -0.037 0.075 0.00172 0.00200 77 Sierra Leone -0.132 0.127 0.00443 0.00375 55 Exp. is the estimated coefficient on experience and Exp2. on experience squared. Std. Err. refers to Standard Errors. The dependent variable is the log of annual earnings. The specification of the regression is described in the text. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple at the 1-percent level. 41 Table 5 Standard Specification (I) (II) (III) (IV) A. All Countries Log Open 1.290*** 0.927** 1.262*** 0.830** (0.292) (0.357) (0.290) (0.362) Log GDP per Capita 0.663 0.722* (0.408) (0.388) Regional Dummies No No Yes Yes R-Square 0.120 0.153 0.239 0.276 # Observations 93 93 93 93 B. Non-Oil Countries Log Open 1.222*** 0.926** 1.189*** 0.850** (0.287) (0.363) (0.286) (0.372) Log GDP per Capita 0.546 0.572 (0.409) (0.389) Regional Dummies No No Yes Yes R-Square 0.116 0.140 0.233 0.258 # Observations 87 87 87 87 C. Non-OECD/Non-Oil Countries Log Open 0.838** 0.887** 0.729** 0.830** (0.333) (0.395) (0.329) (0.415) Log GDP per Capita 0.275 (0.446) Regional Dummies No No Yes Yes R-Square 0.060 0.061 0.107 0.138 # Observations 67 67 67 67 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. Standard errors are in parentheses and are calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent level, double at the 5-percent and triple at the 1-percent level. 42 Table 6 Average Years of Schooling (I) (II) (III) (IV) (V) A. All Countries Log Open 1.262*** 0.910** 0.788* 0.796* 0.647 (0.290) (0.428) (0.427) (0.437) (0.425) Av. Yrs Schooling 0.334* 0.335* 0.211 0.048 (0.180) (0.195) (0.212) (0.317) Log GDP per Capita 0.517 0.911 (0.472) (0.610) Regional Dummies Yes No Yes No Yes R-Square 0.239 0.206 0.277 0.213 0.303 # Observations 78 78 78 78 78 B. Non-Oil Countries Log Open 1.189*** 0.952** 0.853* 0.825* 0.677 (0.286) (0.447) (0.446) (0.437) (0.430) Av. Yrs Schooling 0.320 0.298 0.149 0.063 (0.199) (0.222) (0.286) (0.330) Log GDP per Capita 0.647 0.854 (0.623) (0.658) Regional Dummies Yes No Yes No Yes R-Square 0.233 0.202 0.28I 0.211 0.296 # Observations 74 74 74 74 74 C. Non-OECD/Non-Oil Countries Log Open 0.729** 0.905* 0.788* 0.822* 0.540 (0.329) (0.504) (0.472) (0.484) (0.429) Av. Yrs Schooling 0.065 0.093 -0.042 -0.167 (0.296) (0.295) (0.389) (0.423) Log GDP per Capita 0.456 1.147 (0.738) (0.933) Regional Dummies Yes No Yes No Yes R-Square 0.107 0.080 0.118 0.081 0.144 # Observations 54 54 54 54 54 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. Standard errors are in parentheses and are calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent level, double at the 5-percent and triple at the 1-percent level. 43 Table 7 Share of Manufacturing and GADP (I) (II) (III) (IV) (V) A. All Countries Log Open 1.021** 0.821** 0.720** 0.465 0.833** (0.402) (0.369) (0.279) (0.306) (0.335) Log GDP per Capita 0.783* 0.785* -0.602 -0.055 -0.598 (0.443) (0.425) (0.439) (0.481) (0.450) Share Manufacture 0.054** 0.069** 0.028 (0.027) (0.028) (0.026) GADP 8.157*** 7.747*** 7.671*** (1.624) (1.714) (1.468) Regional Dummies No Yes No Yes No R-Square 0.222 0.291 0.300 0.343 0.333 # Observations 75 75 91 91 75 B. Non-Oil Countries Log Open 0.988** 0.891** 0.740*** 0.532* 0.897*** (0.409) (0.407) (0.281) (0.303) (0.330) Log GDP per Capita 0.821* 0.759* -1.206*** -0.731 -1.026** (0.442) (0.440) (0.383) (0.484) (0.435) Share Manufacture 0.048 0.063** 0.035 (0.029) (0.031) (0.029) GADP 10.325*** 9.878*** 9.374*** (1.6434) (1.841) (1.623) Regional Dummies No Yes No Yes No R-Square 0.218 0.315 0.334 0.363 0.346 # Observations 72 72 85 85 72 C. Non-OECD/Non-Oil Countries Log Open 0.954** 0.771* 0.739** 0.512* 0.816** (0.451) (0.428) (0.291) (0.312) (0.331) Log GDP per Capita 0.251 0.898* -1.224*** -0.734 -0.980** (0.543) (0.516) (0.394) (0.501) (0.465) Share Manufacture 0.048 0.064 0.012 (0.039) (0.040) (0.036) GADP 10.363 10.131*** 10.196*** (2.188) (2.401) (2.093) Regional Dummies No Yes No Yes No R-Square 0.048 0.208 0.271 0.24 # Observations 53 53 65 53 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. Standard errors are in parentheses and are calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent level, double at the 5-percent and triple at the 1-percent level. 44 Table 8 Investment per Worker (I) (II) (III) (IV) (V) (VI) A. All Countries Log Open 0.773* 0.310 0.819** 0.547 0.558* 0.187 (0.402) (0.384) (0.399) (0.402) (0.332) (0.351) Log Inv. Per Worker 0.828** 1.506*** 1.218 1.835 0.138 0.674 (0.401) (0.483) (1.061) (1.124) (0.494) (0.529) Log GDP per Capita -0.646 -1.430 (1.313) (1.326) GADP 6.100*** 4.558** (1.565) (1.919) Regional Dummies No Yes No Yes No Yes R-Square 0.241 0.384 0.246 0.387 0.332 0.381 # Observations 82 82 82 82 82 82 B. Non-Oil Countries Log Open 0.773* 0.330 0.829** 0.585 0.583* 0.340 (0.407) (0.390) (0.405) (0.408) (0.335) (0.339) Log Inv. Per Worker 0.840** 1.493*** 1.278 1.877* 0.099 0.907 (0.405) (0.484) (1.074) (1.131) (0.535) (0.717) Log GDP per Capita -0.727 -1.525 (1.341) (1.349) GADP 6.303*** 4.707* (1.917) (2.430) Regional Dummies No Yes No Yes No Yes R-Square 0.250 0.377 0.256 0.384 0.330 0.399 # Observations 78 78 78 78 78 78 C. Non-OECD/Non-Oil Countries Log Open 0.636 0.378 0.767* 0.471 0.542 0.381 (0.431) (0.437) (0.445) (0.442) (0.354) (0.409) Log Inv. Per Worker 0.638 1.192** 1.831* 2.150* 0.131 0.899 (0.492) (0.576) (1.071) (1.192) (0.559) (0.748) Log GDP per Capita -2.345 -1.940 (1.424) (1.485) GADP 6.772*** 2.949 (2.061) (3.364) Regional Dummies No Yes No Yes No Yes R-Square 0.137 0.287 0.199 0.328 0.212 0.297 # Observations 58 58 58 58 58 58 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. Standard errors are in parentheses and calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple at the 1-percent level. 45 Table 9 Imports and Exports per Worker (I) (II) 1970 1975 1980 1970 1975 1980 A. Non-Oil Countries Log Open 1.157*** 1.141*** 1.141*** 0.743** 0.755** 0.891*** (0.413) (0.363) (0.363) (0.353) (0.330) (0.331) Log Imports per Worker 0.569* 0.601** 0.773** 0.301 0.587** 0.591 (0.299) (0.315) (0.298) (0.284) (0.332) (0.388) Log MNF Imp from OECD 0.484 0.564** 0.637** 0.209 0.493 0.457 (0.307) (0.319) (0.277) (0.243) (0.312) (0.322) Log MNF Imp from Non-OECD 0.506** 0.356 0.594** 0.069 0.066 0.318 (0.232) (0.260) (0.297) (0.315) (0.255) (0.337) Log of Computer Imp per Worker 0.437*** 0.399*** 0.520*** 0.273 0.088 0.115 (0.165) (0.150) (0.136) (0.242) (0.250) (0.256) Log Exports per Worker 0.690*** 0.776*** 0.842*** 0.296 0.573** 0.807** (0.250) (0.233) (0.228) (0.309) (0.312) (0.333) Log MNF Exp to OECD 0.522*** 0.559** 0.623*** 0.339** 0.23 0.323 (0.135) (0.223) (0.220) (0.143) (0.340) (0.324) Log MNF Exp to Non-OECD 0.520*** 0.578** 0.794*** 0.222 0.412 0.344 (0.176) (0.223) (0.214) (0.198) (0.252) (0.341) # Observations 71 74 74 65 69 72 C. Non-OECD/Non-Oil Countries Log Open 1.006** 1.019** 1.019** 0.618* 0.653** 0.841** (0.467) (0.404) (0.404) (0.358) (0.312) (0.335) Log Imports per Worker 0.473 0.521 0.733** 0.191 0.538 0.65 (0.335) (0.339) (0.330) (0.287) (0.346) (0.420) Log MNF Imp from OECD 0.430 0.547 0.654** 0.173 0.551 0.590* (0.349) (0.347) (0.308) (0.253) (0.340) (0.350) Log MNF Imp from Non-OECD 0.406 0.209 0.496 -0.104 -0.133 0.195 (0.259) (0.293) (0.333) (0.381) (0.284) (0.370) Log of Computer Imp per Worker 0.430** 0.366** 0.497*** 0.117 0.115 0.344* (0.187) (0.174) (0.149) (0.259) (0.256) (0.201) Log Exports per Worker 0.546* 0.683** 0.771*** 0.083 0.422 0.762** (0.304) (0.270) (0.258) (0.335) (0.309) (0.354) Log MNF Exp to OECD 0.464*** 0.492** 0.575** 0.334** 0.235 0.384 (0.151) (0.217) (0.219) (0.155) (0.388) (0.371) Log MNF Exp to Non-OECD 0.457** 0.505** 0.719** 0.146 0.322 0.323 (0.196) (0.240) (0.243) (0.209) (0.276) (0.379) # Observations 52 55 55 46 50 53 Dependent variable is the estimated return to home country experience. All specifications include significant regional dummies. (II) additionally controls for log of GDP per capita, share of manufacturing and government antidiversion policy (GADP). Estimation technique is OLS. Standard errors are in parentheses and are calculated using the White estimator. MNF refers to manufacturing. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple at the 1-percent level. 46 Table 10 Self-Selection Migration rate conditional on: age education (I) (II) (I) (II) A. All Countries Log Open 0.851** 0.742* 0.950** 0.473 (0.424) (0.453) (0.416) (0.314) Log GDP per Capita 0.793** -0.162 0.659 -0.655 (0.375) (0.406) (0.596) (0.482) GADP 4.712*** 8.767*** (1.527) (1.755) Share Manufacture 0.047 0.044 (0.032) (0.032) R-Square 0.388 0.458 0.226 0.438 # Observations 80 68 59 54 B. Non-Oil Countries Log Open 0.870** 0.651 0.963** 0.465 (0.434) (0.432) (0.414) (0.311) Log GDP per Capita 0.670* -0.956* 0.656 -0.688 (0.383) (0.505) (0.591) (0.486) GADP 9.283*** 8.935*** (1.853) (1.895) Share Manufacture 0.040 0.045 (0.032) (0.033) R-Square 0.371 0.437 0.223 0.432 # Observations 74 65 57 52 C. Non-OECD/Non-Oil Countries Log Open 0.800* 0.561 0.744* 0.224 (0.470) (0.465) (0.428) (0.305) Log GDP per Capita 0.604 -0.924* -0.152 -0.618 (0.464) (0.520) (0.759) (0.570) GADP 9.402*** 8.986*** (2.477) (2.595) Share Manufacture 0.029 0.028 (0.036) (0.044) R-Square 0.260 0.311 0.064 0.025 # Observations 55 47 38 34 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. Migration rates are added as regressors in the first step regression. Standard errors are in parentheses and are calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple a the 1-percent level. 47 Table 11 Returns to Experience, Education and English Proficiency OLS Fixed Effects Exp 0.022*** 0.020*** (0.003) (0.001) Interaction with Exp 0.019*** 0.019*** (0.005) (0.002) Exp2 / 100 -0.041*** -0.037*** (0.006) (0.002) Interaction with Exp2/100 -0.046*** -0.042*** (0.009) (0.005) R-square 0.688 0.320 # Observations 173137 173137 Dependent variable is the log of annual earnings. Regressions include dummies for year of education, marital status, English speaking status, health status, SMSA, census division, 1980 and year of immigration, as well as interaction terms between the 1980 dummy and all other dummies. All experience terms are interacted with a dummy that assumes value one if the individual has at least 12 years of education. Standard errors are in parentheses. They are robust and account for clustering within country of origin. Single asterisk asterisks denotes statistical significance at the 10-percent, double at the 5-percent, triple at the 1-percent level. 48 Table 12 Returns to Experience and Education Home country skill US immigrants skill Highly skilled Low skilled distribution distribution (I) (II) (III) (IV) A. All Countries Log Open 0.929** 0.953 0.909** 0.855* (0.430) (0.877) (0.413) (0.437) Log GDP per Capita -1.071* -0.349 -1.123** -0.895 (0.562) (1.035) (0.532) (0.586) GADP 6.087*** 7.068* 6.468*** 5.874*** (2.250) (3.878) (1.712) (1.554) Share Manufacture 0.069** 0.046 0.066** 0.074** (0.031) (0.058) (0.031) (0.031) R-Square 0.484 0.176 0.588 0.573 # Observations 66 66 66 66 B. Non-Oil Countries Log Open 0.905** 1.159 0.867** 0.760* (0.444) (0.884) (0.413) (0.433) Log GDP per Capita -1.104 -1.137 -1.280** -0.931 (0.711) (1.186) (0.631) (0.736) GADP 6.243* 9.999** 7.379*** 7.234*** (3.330) (4.547) (2.550) (2.527) Share Manufacture 0.073** 0.062 0.064** 0.063* (0.033) (0.059) (0.031) (0.035) R-Square 0.538 0.202 0.586 0.573 # Observations 63 63 63 63 C. Non-OECD/Non-Oil Countries Log Open 0.281 0.427 0.443 0.199 (0.437) (0.974) (0.390) (0.402) Log GDP per Capita -0.840 -0.627 -8.779 -0.478 (0.699) (1.060) (0.667) (0.680) GADP 9.305*** 11.498* 5.694** 6.485** (2.206) (5.837) (2.295) (2.448) Share Manufacture -0.038 -0.034 0.041 0.014 (0.033) (0.091) (0.035) (0.035) R-Square 0.291 0.115 0.462 0.462 # Observations 44 44 44 44 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. Standard errors are in parentheses and are calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple a the 1-percent level. 49 Table 13 Returns to Experience and English speaking Origin English not widely spoken English widely spoken (I) (II) A. All Countries Log Open 1.180** 1.372** (0.533) (0.510) Log GDP per Capita 0.025 -2.094** (0.429) (0.829) GADP 2.718* 13.372*** (1.602) (2.666) Share Manufacture 0.053 0.147** (0.037) (0.070) R-Square 0.536 0.524 # Observations 51 24 B. Non-Oil Countries Log Open 1.180** 1.372** (0.547) (0.510) Log GDP per Capita 0.333 -2.094** (0.636) (0.829) GADP 1.176 13.372*** (2.907) (2.666) Share Manufacture 0.053 0.147** (0.039) (0.070) R-Square 0.530 0.524 # Observations 48 24 C. Non-OECD/Non-Oil Countries Log Open 1.209** 1.630** (0.549) (0.634) Log GDP per Capita 1.002 -2.578** (0.668) (1.131) GADP -0.634 11.725*** (3.232) (2.715) Share Manufacture 0.074* 0.185* (0.040) (0.099) R-Square 0.547 0.541 # Observations 34 19 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. All specifications include significant regional dummies. Standard errors are in parentheses and are calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple a the 1-percent level. 50 Table 14 Without Experience in Home Country (I) (II) (III) (IV) A. All Countries Log Open 0.012 0.125 0.531 0.021 (0.630) (0.655) (0.617) (0.776) Regional Dummies No Yes Yes Yes R-Square 0.000 0.168 0.184 0.192 # Observations 70 70 70 58 B. Non-Oil Countries Log Open -0.014 0.932 0.475 0.0093 (0.635) (0.660) (0.625) (0.780) Regional Dummies No Yes Yes Yes R-Square 0.000 0.109 0.122 0.109 # Observations 66 66 66 58 C. Non-OECD/Non-Oil Countries Log Open 0.080 0.192 0.574 -0.014 (0.810) (0.764) (0.727) (0.776) Regional Dummies No Yes Yes Yes R-Square 0.000 0.118 0.133 0.117 # Observations 50 50 50 40 Dependent variable is the estimated return to home country experience in percent. Estimation technique is OLS. All specifications include significant regional dummies. (III) controls additionally for GDP per capita and (IV) GADP and Share of Manufacturing. Standard errors are in parentheses and are calculated using the White estimator. Single asterisk denotes statistical significance at the 10-percent, double at the 5-percent and triple a the 1-percent level. 51 Figures Figure 1 Log Earnings/Experience Profile Open= 0.2 Open= 0.3 s 9.4 Earning 9.2 9 Annual of 8.8 Log 8.6 1 4 7 10 13 16 19 22 25 28 31 34 37 40 Years of Experience e 0.03 0.02 Experienc 0.01 to 0 Returns -0.01 0 3 6 9 12 15 18 21 24 27 30 33 36 39 Years of Experience 52 Figure 2 Returns to Experience and Years of Experience Korea Taiwan 0.04 0.03 0.02 Experience 0.01 to 0 -0.01 Returns-0.02 1 4 7 10 13 16 19 22 25 28 31 34 37 40 Years of Experience Belgium Portugal 0.1 0.05 Experience 0 to -0.05 Returns -0.1 1 4 7 10 13 16 19 22 25 28 31 34 37 40 Years of Experience 53