Import penetration and returns to tasks: recent evidence from the Peruvian labour market

On average the country annually grew by about 6.5% between 2002 and 2008 and, in the aftermath of the recent global crisis, it kept on growing at a similar pace (World Bank 2015).

The weight of exports on GDP rose from 16% in 2001 to 24% in 2013 and Peru’s average MFN rate declined from 10.7% in 2004 to 3.4% in 2013 over the same period, one of the lowest in the continent. Consequently, imports rose from 18% to 26% of GDP over the same period (World Bank 2015).

Namely, from the maximum peak of 18% in 2007 to 15% in 2013 (World Bank 2015).

Muendler (2008) explores the evolution of employment in the aftermath of Brazil’s trade reforms and documents for Brazil a marked occupation downgrading and a simultaneous education upgrading within the traded sector of Brazil. Also, exporters experienced a severe employment downsizing. Currie and Harrison (1997) examined the employment effects of trade reforms in Morocco and similarly found that, although manufacturing employment was, on average, unaffected, exporters and highly exposed firms were characterised by significant employment losses.

Our work is also near to that of Amiti and Davis (2011) who study the effect of trade liberalisation on manufacturing wages in Indonesia. Using data from 1991 to 2000, the authors test their model on the impact of final and intermediate input tariff cuts on workers’ wages and find that the effects are heterogeneous depending on how firms engage in the global economy. They, nonetheless, do not distinguish between skilled and low-skilled workers. For the same country Lee and Wie (2015b) show that the diffusion of foreign technologies through imports and foreign direct investment caused demand to shift towards more skilled labour and increased wage inequality.

Namely, \({ EXP}^{{ US}}_{jt}\) = (us exports of sector j going to non-lac)/(us production of sector j). Then \({ EXP}^{{ US}}_{kt}\) is computed according to Equation (10) in Sect. 4.

Nevertheless, in the robustness checks we use alternative IVs to further control for the potential existence of common demand shocks.

In a simple setting, workers can choose between two occupations, say white versus blue collar jobs, and are characterised by different skill endowments, motor and cognitive skills. Workers will self-select into the occupation returning the highest premium to the skill they are endowed with. For instance, cognitive-skilled workers choose to work as white collars if cognitive skills are more rewarded in white collar jobs.

In the present setting we characterise occupations by a one-dimensional task measure and consider the sorting of workers into a continuum of occupations, which extends the traditional binary choice Roy model, as opposed to the general task approach in which occupations are described by a bundle of task measures.

It can be argued that information about parents’ educational attainment and occupation could be used as sources of exogenous variation to identify the effects of the tasks performed on wages, similarly to the instrumental variable approach to the estimation of the Mincerian returns to education. However, the survey does not contain information on the family of origin and limiting the analysis to offsprings would considerably restrict the sample.

Although the ENAHO survey has some information on individuals’ labour conditions, it has no information on their occupation of employment. The share of informal workers was only available at the 3-digit ISIC Rev. 3 sector level and has been kindly provided by Ana Cisneros Acevedo.

Also, informal workers primarily work in agriculture, rather than in manufacturing and in non-urban areas (CEPLAN 2016). In the Lima metropolitan area—the geographical location of workers in our sample—informal workers account for 56% of total employment in our sample period (WIEGO 2010) compared to roughly 70% for non-agricultural employment in the rest of the country (ILO 2014).

We provide results for the two-step estimator based on uncorrected cluster standard errors as a robustness check.

Notice that under the null hypothesis of exogeneity, standard errors need not to be corrected for the presence of generated regressors. In this respect, the procedure of testing for exogeneity is similar to that described by Wooldridge (2010) when illustrating the Hausman test in the presence of heteroskedasticity.

We collect data up to 2011. In 2010 classification of sectors of activity changed from isic Rev. 3 to isic Rev. 4 and, due to the lack of a proper fit between the two classifications, 2010 and 2011 are not considered in this work.

From the initial sample we only retain workers working in the private sector and therefore drop those working in electricity, gas, steam and hot water supply; collection, purification and distribution of water; public administration and defence and compulsory social security.

Our main results are robust to the inclusion of workers working in these sectors. Results are not shown for the sake of brevity, but they are available upon request.

Actually, own calculations on wits-comtrade data reveal that capital goods and industrial supplies represent about 75% of non-fuel and non-primary imports. Furthermore, Fig. 7 in “Appendix A” shows that a large, although slightly declining, share of manufacturing imports originate from the eu and the usa At the same time, China is gaining increasing importance to the detriment of the Latin American neighbouring countries. In particular, Peru mostly imports capital goods and industrial supplies from the eu, usa and China. Purchases from the latter country, though, are also characterised by a non-negligible share of final goods (roughly 20%).

The unido Industrial Statistic Database (indstat4) contains highly disaggregated data on the manufacturing sector. It provides, among other, information on output levels up to the 4-digit level of the isic Rev. 3. More information on how to access the database is available at the following link: http://​www.​unido.​org/​resources/​statistics/​statistical-databases.​html.

Although epe provides the 4-digit isic code of sectors where individuals are employed, trade flows from wits-comtrade are only available at the 3-digit level. For this reason, we attribute the same trade and output flows to individuals working in different 4-digit sectors who, nonetheless, belong to the same 3-digit sector. The slightly more aggregate sector detail has the advantage of producing sector cells with a not too small number of individuals.

It was not possible to compute import penetration measures for the primary and service sectors. In both cases detailed information on sectors’ output to be included at the denominator of the formula was not available. Also, as far as service sectors are concerned, it is worth highlighting that most of these sectors are non-tradable and for those few tradable service sectors detailed information on imports is unavailable.

In this respect, Handel (2012) shows that uso*net indicators are highly correlated with occupation-level measures of job characteristics for a set of advanced economies from the European Social Survey and the International Social Survey Programme. Also, o*net-based cognitive and physical skill measures highly reflect the education level by occupation for countries in the European Labour Force Surveys, among which a few middle-income economies are present. We were prevented from running a similar comparison for Peru on the basis of our data, due to the lack of reliable information on years of schooling.

Potential lack of comparability notwithstanding, our empirical setting accounts for the measurement error in MS, which is addressed via the kv-iv estimation approach.

As an example, consider the o*net abilities “Manual dexterity” and “Fluency of Ideas”. Raw o*net scores for both items range from 1 to 5. In a sample of us workers we can normalise scores by dividing them for the maximum value. For “Bookbinders and related workers” (ISCO88 code \(=\) 7345) the raw score of “Manual dexterity” equals 3.44 which corresponds to 0.688 when rescaled, while the score for “Fluency of Ideas” is 2.185, corresponding to a rescaled value of 0.437. When normalising raw scores according to the distribution of workers across occupations in the Peruvian labour market “Manual dexterity” takes the value 0.395, meaning that almost 40% of the working population has an intensity of manual dexterity lower than that of bookbinders, which makes it a fairly manual-intensive job. At the same time, “Fluency of Ideas” takes the value 0.067 meaning that only 6.7% of the population has an intensity of fluency of ideas lower than that of bookbinders, which makes it a low cognitive-intensive occupation. By accounting for the peculiarity of the Peruvian labour market, this job requires more “Manual dexterity” than in the usa.

More details on exact correlations between our task indicators and those used in the literature are available in “Appendix A”.

In all the specifications presented in Table 5 and 6, individual-level controls are suppressed from the tables for the sake of brevity. Nonetheless, complete results for the baseline specifications and for first-stage estimation results are available in Tables 11, 12 and 13 in “Appendix B”.

Calculations are as follows: from coefficients estimates in Column [2]: \((0.139-0.109*{ MS})*\Delta { IP}^{09/04}_{j}\) with \({ MS}=0.27\) for engineers and \(\Delta { IP}^{09/04}_{j}=0.12\).

Calculations are as follows: from coefficient estimates in Column [6]: \((0.752-1.026*{ MS})*\Delta { IP}^{09/04}_{k}\) with \({ MS}=0.27\) and \(\Delta { IP}^{09/04}_{k}=-0.02\) for engineers and \({ MS}=0.85\) and \(\Delta { IP}^{09/04}_{k}=0.20\) for assemblers.

If this were the case, however, this would imply that industry-/occupation-specific human capital is less relevant for highly cognitive-intensive activities. Ebenstein et al. (2014), instead, for routine workers in the usa switching out of manufacturing, interpret wage losses as stemming from the loss of industry and, especially, occupation-specific tenure.

Biscourp and Kramarz (2007) on French firms document that imports of both intermediates and final goods are positively related to skill upgrading within firms and industries.

More specifically, from WITS-COMTRADE we retrieved data on Peruvian import flows at 6-digit HS 2002 and jointly reclassified these product-level import flows according to the BEC and to the ISIC Rev. 3 classifications by means of readily available crosswalks from UNCTAD and RAMON websites. As a consequence, we were able to dissect, from the bulk of imports referable to an ISIC sector, import flows pertaining to each of the BEC categories. We classified as capital imports of capital goods (BEC one-digit code 4) and processed industrial supplies (BEC code 22). Remaining imports are mostly made up of final goods, as the incidence of primary goods is negligible or even non-existent in industrial sectors.

http://​faculty.​som.​yale.​edu/​peterschott/​sub_​international.​htm.

In this regard, we follow Autor et al. (2003), who build a measure of computerisation as the sum of capital stocks of private non-residential fixed assets by industry. In particular, they define the computer stock as the sum of stocks in mainframe computers, personal computers, packaged and custom software, printers, terminals, storage devices and other computers and peripheral equipment manufacturing. All stock variables are measured in constant 2009 dollars and classified according to the BEA industry classification which is directly comparable to the 1997 North American Industry Classification System (NAICS).