Research Paper Series Globalisation and Labour Markets International Competition, Returns to Skill and Labor Market Adjustment the Centre Acknowledges Financial Support from the Leverhulme Trust under Programme Grant F114/bf Acknowledgements We Are Grateful to Participants at the Midwest Internation

The finding, interpretations, and conclusions expressed in this paper are entirely those of the authors and do not necessarily reflect the views of the World Bank. Abstract This paper examines whether increased import competition induces domestic workers to skill upgrade and/or switch industries. The analysis makes use of a large unique longitudinal matched employer-employee dataset that covers virtually all workers and firms in Portugal over the 1986-2000 period. Our identification strategy uses two exogenous changes in the degree of international competition. First, we exploit the strong appreciation of the Portuguese currency in 1989-1992 and pre-existing differences in trade exposure across industries in a differences-indifferences estimation. Second, we make use of changes in industry-specific (source-weighted) real exchange rates. A bivariate probit model is used to analyse the impact of increased international competition on skill-upgrading and/or industry switching. Based on both empirical strategies, and on two different skill definitions, we find strong confirmation for the hypothesis that increased international competition increases the returns to skill and induces skill upgrading.


Introduction
In recent decades the increased integration of national product markets has stimulated a large literature aiming to identify and explain its labour market consequences. Among the consequences of particular interest are adjustments in the market returns to different skills/occupations and changes in workers skill-acquisition decisions. The former were the early focus of this literature, in particular whether trade liberalisation was an important driver of the increased wage inequality between skilled and unskilled workers observed in many high income countries. The general consensus from this seems to be that it may have contributed to a rise in the skill premium but played a small role relative to skilled biased technological change (Slaughter 2000, Acemoglu 2002, Machin 2003. 1 Other recent contributions argue that organisational change (Caroli and Van Reenen, 2001;Black and Lynch, 2004;Garicano and Rossi-Hansberg, 2006) and declining unionisation (Machin, 1997;Card, 2001) have also played an important role.
A recent paper by Guadalupe (2007) reports a previously unnoticed driver of the increase in returns to skill: changes in the degree of product market competition. Guadalupe identifies a mechanism whereby an increase in competition within an industry induces a rise in returns to skill. If product markets are imperfectly competitive, greater competition increases the sensitivity of profits to production costs. Provided that skilled workers are more productive than unskilled workers, an increase in competition raises the sensitivity of profits to the proportion of skilled workers hired.
This induces a rise in demand for skills, which translates into higher returns. Using a panel of UK workers, she finds empirical support for this hypothesis. Exploiting two quasi-natural experiments which affected different sectors in different periods (a sharp appreciation of Sterling and implementation of the European Single Market Program), Guadalupe identifies a causal relationship between implied changes in the degree of competition and returns to skill within each industry.
Our paper makes two contributions. We begin by providing further evidence on the impact of within industry changes in competition on returns to skill. To do so we exploit a large longitudinal matched employer-employee dataset for Portugal that covers almost all workers and firms in the Portuguese private sector over the 1986-2000 period, merged with very detailed trade data disaggregated at the industry level by both country of origin of imports and destination of exports.
Our identification strategy involves two exogenous measures of changes in international competition. First, following Cuñat and Guadalupe (2005) and Guadalupe (2007), we exploit a 1 An important exception to this consensus is Wood (1998). For recent surveys of literature on globalization and inequality see Greenaway and Nelson (2002), Feenstra and Hanson (2003), Bardhan (2005) and Goldberg and Pavcnik (2007). 1 strong appreciation of the Portuguese currency in 1989-1992 (over 25%) and pre-existing differences in cross-industry trade exposure in a differences-in-differences estimation. Second, following Revenga (1992), Campa and Goldberg (2001) and Bertrand (2004), we make use of industry-specific real exchange rates, for which we have data for a later period (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000). Based on both strategies, and on two different skill definitions, we find strong confirmation for the hypothesis that within industry increases in international competition are an important determinant of rising wage inequality.
The second, and perhaps most important, contribution is that we investigate whether changes in competition also impact on skill upgrading. Specifically, we analyse whether a within industry increase in foreign competition causes skill upgrading and industry switching. This reallocation may result from workers' reactions or firms' decisions. The analysis of worker reactions relates to the theoretical framework in Falvey et al (2007), where we argue that skill acquisition is an important part of the adjustment process to a change or shock that impacts on relative wages. Here we also focus on adjustment by the existing workforce by explicitly considering reallocation between industries and skills. However, we focus on a different underlying mechanism for the change in relative wages: whereas in Falvey et al (2007) the driver was the change in relative output prices that followed liberalisation here its main cause is a within industry increase in competition. 2 Using the same dataset and identification strategies, we estimate transition probabilities between skills and/or industries (controlling for worker, firm and sector characteristics) in a bivariate probit specification. We distinguish between four alternatives: no change; moving industry; skill upgrading; moving industry and skill upgrading. Our empirical results strongly support the view that labour market adjustment to globalisation involves significant worker movements across sectors and skills. We find that increased international competition induces skill acquisition and decreases skill downgrading.
Understanding the implications of increased competition for skill acquisition is important given the theoretical possibility of poverty traps generated by lack of education (Barham et al. 1995) and occupational choice (Banerjee and Newman 1993), and the role of human capital accumulation in growth. Additionally, uncovering the channels influencing skill upgrading in an open economy context may shed light on how human capital accumulates as countries grow and what policies are more effective in expediting this process.
2 Modelling worker transitions induced by a trade shock is also the focus of Davidson and Matusz (2000, 2002 and Long et al. (2007). However, whereas the first focuses on consequences of industry specific human capital for the adjustment process, the second focuses on firm specific human capital.

2
The remainder of the paper is organised as follows. In Section 2 we outline the empirical methodology. In Section 3 we describe the data. Section 4 presents some descriptive statistics about workers' transitions. Section 5 discusses the empirical estimates, and Section 6 concludes.

The effect of increased international competition on the returns to skill
To identify an exogenous change in the degree of product market competition, Guadalupe Abowd and Kramarz (1999) and Abowd et al (2006), we argue that, given the existence of heterogeneity in compensation and retention policies adopted by firms, the estimated wage equation should ideally control for worker and firm effects. With regard to firm characteristics, our data includes information on size, labour productivity, age, proportion of foreign owned capital, location and industrial concentration. To account for both worker and firm unobserved effects, we exploit the longitudinal nature of our matched worker-firm data and estimate models with worker and spell (worker-firm) fixed effects.
While in individual fixed-effects models the effect is identified out of individuals who stay in the same firm as well as individuals who move after a shock, in spell fixed-effects models the effect is identified out of variation over the time period in which the worker is employed in a given firm, thereby ensuring that unobserved changes to the industry composition of employment are not driving the results.

3
A second important innovation is that we are able to use a more informative definition of skill level.
In Guadalupe's study skills evaluation was exclusively based on workers' occupations; our data permits evaluation based on occupations and schooling levels. Given the focus on returns to skill, it is particularly important to investigate whether results are robust to these different measures.
Another innovative feature is that, besides the 1989-1992 exchange rate appreciation, we use an additional empirical strategy to identify an exogenous change in competition: changes in industry source-weighted real exchange rates. This is computed using real exchange rates and bilateral trade data on imports for each industry. This strategy was established in the literature by Revenga (1992) to investigate the effect of international competition on industry wages and employment in US manufacturing, and has been recently adopted by Campa and Goldberg (2001), Bertrand (2004) and Cuñat and Guadalupe (2006) to examine the effect of competition on, respectively, industry wages and employment, the sensitivity of wages to the unemployment rate, and provision of incentives to top managers inside the firm.
We start with the following baseline Mincerian wage equation: where is the wage of worker i employed at firm J of industry K at year t; iJKt w Jt Z is a vector of characteristics of firm J at year t (size, labour productivity, age, proportion of foreign owned capital, regional dummy); Kt hhi is the Herfindahl-Hirschman index for industry in which firm operates; K J J ϕ is a firm unobserved effect and the other variables are as above 3 . The main parameter of interest is λ , which reflects how returns to skill vary with international competition and we expect it to be positive.
We consider two different specifications of equation (2). In the first, the worker's level of skill is based on schooling, defined as the number of completed years of education (equation (2.1)). In the second, skill is computed using the International Labour Office's (ILO) correspondence Table   between major groups of occupations in the 1988 International Standard Classification of Occupations (ISCO-88) and skill level (equation (2.2)). The ISCO-88 "is based on the nature of the skills required to carry out the tasks and duties of the job not the way these skills are acquired" (Hoffmann 2000, pp. 2), considering formal education, training and experience.  Elias et al (1999) 4 . In the econometric analysis we aggregate skill groups 1 and 2 as low skilled workers, which is then used as the base category in the estimation of returns to skill. Normally requires a degree or an equivalent period of relevant work experience.
Requires a body of knowledge associated with a period of postcompulsory education but not to degree level.
Source: International Labour Office (1990, pp. 2-3) and Elias et al. (1999) as in Upward and Wright (2003 Appreciation of the home currency leads to increased international competition for national firms for two reasons: it reduces the prices that foreign competitors can offer in the home country market; and it encourages more entry by increasing the number of potential foreign firms that can sell in the home country. As argued by Revenga (1992), Bertrand (2004) and Guadalupe (2007), exchange rate movements are largely unpredicted and exogenous to the behaviour of firms and workers within each industry. We adopt two complementary empirical strategies. First, we identify an episode of sharp appreciation in the Portuguese currency that can be regarded as a quasi-natural experiment that affected sectors differently. Figure 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Source: Bank of Portugal.
In Figure 2 we plot the difference in mean log wages between skilled and unskilled workers of the manufacturing sector over our sample period. Whether we identify skill by schooling only or using the ILO's skill definition, between 1989 and 1992 this difference increased sharply.  We use the pre-and post-appreciation period (1986-1988 and 1989-1992 respectively) and preappreciation differences between sectors in trade openness to identify exogenous changes in international competition in a difference in differences specification. The advantage of this is that it controls for pre-existing differences across industries and changes common to all industries. The hypothesis is that the appreciation represents a higher increase in the degree of product market competition in the sectors that are (ex-ante) more open (treatment group) relative to those fairly closed (control group). More specifically, we estimate: where is the degree of openness to trade of industry K indop88 K and t post89 is a dummy variable that takes the value zero in the pre-appreciation period (1986)(1987)(1988) and one during the appreciation (1989)(1990)(1991)(1992). The rest of the variables are defined as before. Note that since trade openness may vary endogenously with real appreciation, the variable is computed as the level of openness for 1988 K indop88 5 . In this specification λ is the differences-in-differences estimate of returns to skill and captures how these vary with international competition. To get this estimate it is necessary to control for differences in returns to skill before and after the experiment (captured by ς ) and differences in returns to skill between sectors with different degrees of openness (captured by ϕ ).
The second empirical strategy uses changes in source-weighted industry real exchange-rates to measure exogenous changes in international competition. This is defined as the weighted average of the log real exchange rates of importing countries, where the weights are the shares of each trade partner in the industry's total imports in a base period. This index varies across industries based on the composition of imports by country of origin and has also been used by Revenga (1992), Bertrand (2004), Cuñat and Guadalupe (2006) to instrument for changes in import penetration. To avoid potential endogeneity issues, the weights should be prior to the period under analysis. Given that the first years for which bilateral data on imports by industry is available are 1990 and 1991, the period under analysis is restricted to 1991-2000. We therefore estimate equation (2), where is now the source-weighted real exchange rate index for industry K IC K .

Increased international competition on skill acquisition and industry relocation
To examine the impact of increased international competition on skill acquisition and industry relocation we estimate a bivariate probit model. The underlying assumption is that moving skill and moving industry are two alternative routes for adjusting to increased competition, but they are related. The first route can be decomposed into skill upgrading and downgrading. The underlying assumption is that increased competition increases returns to skill (tested in this paper) and generates wage differentials between sectors for workers with the same skill, respectively. The latter effect has been established in the literature by Krueger and Summer (1988), who show that an increase in product market competition, by decreasing monopoly rents, decreases the ability of firms to pay higher wages and generates wage differentials between sectors.
We adopt the following baseline two equation system: where and are two observed binary indicator variables driven by the two equation system of latent propensities to move industry ( ) and skill ( ).
is the percentage change in competition in industry K . We also estimate this system with instead of where the observability criteria for the two sets of binary outcomes is ). We can derive the probabilities associated with each and the unconditional probability of moving industry, As in the estimation strategy for the effect of increased competition on returns to skill, here identifying changes in international competition that are exogenous is crucial and we adopt the same identification strategy as above.
The estimated two equation systems are given by: where is the percentage change in the source-weighted real exchange rate index for industry K exchrate Δ K . The other variables are defined as before. 9

The data set
Our data is from a longitudinal matched employer-employee dataset, "Quadros de Pessoal" [QP], collected by the Portuguese Ministry of Employment, which covers virtually all workers and firms in the Portuguese private sector over the 1986-2000 period, around 200,000 firms and more than 2 million workers each year. It provides comprehensive information on worker's demographic characteristics (age, gender, schooling), occupation characteristics (occupational group, professional category, wage, hours worked) and plant tenure, along with employing firm ID codes.
Firm-level characteristics includes sales, number of employees, equity, percentage of foreign capital, geographical location and date of constitution, along with industry code. This code allows us to merge this data set with very detailed trade data, disaggregated at the industry level with imports by country of origin, collected by the Portuguese National Statistics Office (INE). Trade data is only available for manufacturing for the period 1988-2000. As the Portuguese classification of industries has been revised in 1994 to match the NACE-Rev 2, a concordance was needed and the analysis considers 74 manufacturing industries 6 .
There are two important characteristics that guarantee this data set's coverage and reliability. First, data is collected from a compulsory administrative census that the Ministry of Employment runs annually to check the firm's compliance with labour law. Second, and unique to this data base, firms are required by law to make the information provided to the Ministry available to every worker in a public place of the establishment. Another important advantage is its longitudinal nature that results from the fact that a unique identification number is attributed to each worker the first time he enters the data set. This is based on his social security number that does not change through time.
To control for any coding errors, we have performed extensive checks to guarantee the accuracy of the worker data using information on gender, date of birth and maximum schooling level achieved, as described in Appendix B. After these checks, we kept for analysis full-time wage earners, aged between 16 and 65, earning at least the national minimum wage, working in firms operating in

Workers Transitions: Variable Creation
An industry (skill) switcher is an individual that in the subsequent observation is employed in a different industry (has a different skill). The set of worker transitions contains five alternatives: the worker changes industry retaining the same skill; moves up the skill ladder while remaining in the same industry; moves down the skill ladder while remaining in the same industry; moves up the skill ladder and moves industry; moves down the skill ladder and moves industry. Skill classification follows the ILO's skill definition described in Section 2.1 and industry classification follows the NACE-Rev. 2 at the 2-digit level.

Descriptive Statistics
Moving industry and/or skill appear to be important forms of labour reallocation. Around 17 per cent of all worker-year observations are of some kind of switching. Figure 3 provides some detail on (average) rates of switching (expressed as a percentage of total number of observations).
Switching by changing both industry and skill is least common, on average around 3 per cent of all worker-year observations are of this type. Switching in one dimension alone (either skill or industry) is more common. The rate of industry switching conditional on retaining the same skill is similar to that of changing skill conditional on staying in the same industry, on average around 7 per cent of all worker-year observations are of each of these types. Considering each of these forms of reallocation independently the (unconditional) switching rate raises to around 10 per cent (either for industry or skill).
As Figure 3 makes clear, moving up the skill ladder is more common than moving down, but both movements are important -the share of all workers who are upgraders and downgraders is 9 and 3.8 per cent, respectively. The difference is driven by differences in the fraction that moves skill whilst staying in the same industry. Whereas the unconditional rate of industry switching and its components are similar in both cases, the average rate of skill upgrading, conditional on staying in the same industry, is almost twice as much as that of skill downgrading conditional on staying in the same industry. 7 7 Detailed tables of transition patterns across industries are available from the authors on request.

Estimation Results
As described in Section 2, we begin by investigating whether international competition has a direct effect on relative wages, then estimate the impact of increased competition on the probability of switching between skills and/or industries. In manufacturing the annual average share of workers moving industry is 10 percent and skillupgrading 6 percent. Also significant is the change in the real exchange rate (the annual average change is 3 percent) and level of openness (the annual average level is 37 percent). Three other characteristics of Portuguese labour markers are worth noting: high female participation, low level of education and high share of population in very skilled occupations. 8 Note that 1986 is the first year for which the QP dataset is available and the appreciation finishes in 1992. Baldwin (1988) and Dixit (1989) argue that the appreciation may permanently reshape the competitive structure of the product market and therefore the analysis should be restricted to the end of the episode. Moreover, there has been a recession in the Portuguese economy in 1993 that might contaminate the differences in differences estimate. The second strategy requires industry bilateral trade data prior to the period under analysis. This is only available from 1990.  Table 3 presents the estimated results of equations (2.1) and (2.2) using the 1989-1992 appreciation as the exogenous change in international competition. Columns 1-4 present estimates for skill defined as the number of completed years of education, columns 5-8 report estimates for skill using the ILO's skill definition. For each, we start by estimating the difference results (columns 1-2 and 5-6) and proceed by estimating difference in differences results (columns 3-4 and 7-8). The dependent variable in each column is the log real hourly wage. Each regression includes regional and year dummies to control for disparities in the returns to skill across regions and macro-shocks, respectively. To control for unobservable industry characteristics we include a full set of industrydummies: 74 industry fixed-effects. Additionally, we run each specification with individual fixed effects (where the effect is identified out of the within sector variation in competition) and proceed by estimating the models with spell (work-firm) fixed effects. Whereas in the first case identification comes from the within sector variation in competition, in the second it comes from within spell variation, that is, from variation over the period in which the worker is employed in a 13 given firm. This procedure is important as we are interested in estimating the effect of within industry changes in international competition on relative wages.

International competition and the returns to skill
The coefficient on the interaction of skill variables with is always positive and highly significant in all specifications. In addition, one can see that the difference in difference estimates of returns to schooling (columns 3-4) and high skill (columns 7-8) are lower than the difference estimates (columns 1-2 and 5-6, respectively) and that the coefficients associated with the two-way interaction of the skill variables with, inter alia, and are statistically significant. This confirms the importance of accounting for the fact that more open sectors may systematically pay a lower skill premium to start with and that during 1989-1992 returns to skill increased throughout the economy. Controlling for these, results on column 3-4 and 7-8 show that, relative to the control group, the appreciation had a positive and significant effect on the skill premium of the treatment group, than in industries that are relatively shielded. In particular, the estimated coefficients in column 8 indicate that for an industry with average trade openness (0.37), the pre-appreciation return to skill was 4 percent and the post-appreciation return, 15.4 percent.
The full effect of the appreciation is only captured by the estimated coefficient on the interaction between the skill and the experiment variable (0.024 on column 8), which indicates that for an industry with average exposure to trade in 1988, the effect of the appreciation was to increase the differential (in returns to skill) by 0.88 percent (relative to an industry with no trade prior to 1988).  Table 4 also presents estimated results of equations (2.1) and (2.2) but here we identify exogenous changes as source-weighted industry real exchange rate movements. Columns 1-2 present the estimates for skill defined as the number of completed years of education, columns 3-4 report estimates using the ILO's skill definition. Column 1 and 3 present the individual fixed effects estimates and column 2 and 4 the spell (worker-firm) fixed effects estimates. A full set of industry, year and regional dummies is included in all specifications. The coefficient on the interaction of skill variables with the exchange rate index is positive and highly significant in all specifications, confirming that when the exchange rate index increases the impact of skill acquisition on actual wage is higher. Note that, as pointed out by Bertrand (2004), this methodology, by simultaneously including individual fixed effects and industry dummies, estimates the relationship between changes (not absolute values) in international competition and returns to skill within industries.
Our results are therefore very much in line with Guadalupe (2007). Based on two different identification strategies and skill definitions, we find strong confirmation of the hypothesis that increased international competition is an important determinant of rising wage inequality between skilled and unskilled workers.

International competition, skill acquisition and industry relocation
Using bivariate probit models, Table 5 reports the results for the two equation systems ((4.1) and (4.2)) presenting each of the different forms of relocation that may be induced by increased competition. The marginal effects on the probability of each potential outcome (when all variables are held at their mean) are presented 9 . All regressions include industry fixed effects, year and 9 Regression results can be found in Appendix C. regional dummies and standard errors are clustered by individuals (to account for the fact that the individual error term may be autocorrelated). Columns 1-5 present estimates for the latent propensities of moving industry ( ) and moving skill ( ), columns 6-10 report the estimates for the latent propensities of moving industry ( ) and skill-upgrading ( In addition, Columns 6-10 present bivariate probit estimates of the marginal effects of each of the covariates on the following unconditional probabilities: The marginal effect of interest on the variable is always positive and highly significant. Results therefore indicate that the propensity to switch (industry and/or skill) in response to the appreciation is higher in sectors more open to trade. In addition, results in column 1-6 show that the effect of increased competition on skill acquisition and industry relocation is similar (columns 5 and 6). However, the effect on the probability of changing in one dimension only is stronger than the effect on the probability of changing both skill and industry (coefficient in 88* indop apprec 18 columns 2-3 is higher than in 1). Furthermore, results on the second specification (column 6-10), show that increased international competition in the 1989-1992 appreciation also increases significantly the propensity to skill-upgrade, whether remaining in the same industry or changing industry, though the effect on the latter is weaker. In particular, the estimated coefficients on trade exposure, the post-appreciation dummy and their interaction ( In addition, all signs of the coefficients of the control variables are as expected. Worker covariates indicate that moving between skills and industries is significantly less frequent in older workers but the effect of an additional year of age is higher in younger than older workers. High tenured workers also switch industry and/or skill less frequently. Skilled workers tend to be more mobile than unskilled workers. Male workers move more frequently than female workers. Moreover, industry covariates show that a high degree of industry concentration is associated with higher switching rates. Finally, firm covariates indicate that elevated firm size, labour productivity and participation of foreign capital predict lower switching rates, whereas firm's age increases this rate. Table 6 presents bivariate probit estimates using source-weighted industry real exchange rate fluctuations to identify changes in competition 10 . Results are in line with those previously obtained and confirm the association between increased international competition and additional skill acquisition and industry relocation. For a worker who is average on all characteristics, employed in a firm that is average in all characteristics that belongs to an industry that has the average level of the Herfindahl-Hirschman index, a 1 percentage point increase in the exchange rate index is associated with a probability to move industry 0.03 percentage points higher and probability to skill-upgrading by 0.02 percentage points higher. The effect on skill upgrading while remaining in the same industry is more than twice as large as the effect on skill upgrading and moving industry simultaneously.  McFadden's pseudo R2 * significant at 10%; ** significant at 5%; *** significant at 1%. The period of analysis is 1986-1992. Absolute value of t-statistic in parentheses, based on robust standard errors clustered by individual. All specifications include a full set of industry dummies (74 industry fixed-effects), year and regional dummies. Marginal effects are computed at the mean of each variable. The variables age, tenure are divided by 10, and age-square and hhi by 1000.

Conclusions
This paper identifies international competition as a source of increased returns to skill. Using a large panel (1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000) of the Portuguese manufacturing sector with matched employeremployee data, two different skill definitions, and two different identification strategies of exogenous changes in international competition, we show that returns to skill within an industry increase with competition. Our analysis exploits one quasi-natural experiment: a strong appreciation of the Portuguese currency in 1989-1992 and heterogeneity between industries in trade exposure in a differences-in-differences estimation. It also makes use of industry-specific (source-weighted) real exchange rates whose fluctuations represent changes in international competition.
By affecting returns to skill, increasing international competition influences workers' incentives to acquire skills. This paper identifies increasing international competition as a significant determinant of skill upgrading. This result is valid both when workers stay in the same industry and when they move. This finding indicates that skill acquisition is an important part of the adjustment process to a policy change or shock (exogenous to the wage setting conditions within the country) that changes relative wages. Policy attention to the consequences of increased competition for human capital accumulation seems merited.

Appendices APPENDIX A: Variable description
Wage data is the log deflated monthly earnings actually received including the base wage, tenure-related and other regularly paid components. Wages are deflated using the consumer price index.
Worker covariates: Tenure is defined as the number of years with the same firm, schooling as the number of completed years of education and skilled as having a number of completed years of education higher than 12 (that marks the end of high-school in Portugal).

APPENDIX B: Consistency checks to the information on the employer-employee dataset
We have performed a number of consistency checks on the information provided by "Quadros de Pessoal" to guarantee the accuracy of the data used.

a) Elimination of invalid or duplicated worker identification codes in a given year:
According to the Ministry of Employment valid worker's ID codes have 6 to 10 digits. In each year, observations with codes with less than 6 or more than 10 digits were not considered. Observations with duplicated identification codes were also eliminated. These  (b2) Correcting inconsistent data across years by taking information reported over half of the times as correct 12 . Inconsistent values on gender were replaced by the value reported over half of the cases the worker has been observed, provided that the year of birth in that observation is the same as that reported in more than 50 percent of the cases for that worker.
Similar procedures have been implemented for year of birth and schooling. This affected 0.87 percent, 1.77 percent, 6.65 percent and 1.68 percent of the initial panel for gender, year of birth and schooling, respectively. The whole information on a worker has been dropped when inconsistencies persisted after this correction. This restriction led to dropping 8.4 percent, 1.08 percent and 6.28 percent of the observations for gender, year of birth or schooling, respectively. 11 Two observations. 12 Note that this is a more demanding criterion than simply using the modal value as replacement.

24
The estimated results of the two equation systems ((4.1) and (4.2)) presented in Section 4.3 are marginal effects. Appendix Table C.1 displays the regression results instead.

APPENDIX C: Regression results on the effect of international competition on skill acquisition and industry relocation
(b3) Dropping workers with remaining missing data on gender, age or schooling: 0 percent, 0.15 percent, 0.99 percent due to missing data on gender, age and schooling, respectively.
The checked panel included 22,686,298 worker-year observations and 3,525,485 workers.   3858747 * significant at 10%; ** significant at 5%; *** significant at 1%. The period of analysis is 1986-1992 in Column 1-2 and 5-6 and 1992-2000 in Column 3-4 and 7-8. Absolute value of t-statistic in parentheses, based on robust standard errors clustered by individuals. The variables age, tenure are divided by 10, and age-square and hhi by 1000.