WP/19/##

TRADE EXPOSURE, CONSUMPTION INEQUALITY, AND LABOR MARKETS: A STUDY OF PERUVIAN INDUSTRIES AND DISTRICTS

by Mumtaz Hussain, Yothin Jinjarak, Gonzalo Salinas

IMF Working Papers describe research in progress by the author(s) and are published to elicit comments and to encourage debate. The views expressed in IMF Working Papers are those of the author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

Β© 2019 International Monetary Fund WP/##/##

IMF Working Paper

############ Department

Trade exposure, consumption inequality, and labor markets: A study of Peruvian industries and districts

Prepared by Mumtaz Hussain, Yothin Jinjarak, Gonzalo Salinas1

Authorized for distribution by ########

June 2019

IMF Working Papers describe research in progress by the author(s) and are published to elicit comments and to encourage debate. The views expressed in IMF Working Papers are those of the author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

Abstract

We study spillovers of trade exposure to consumption inequality and labor markets in . Dependent variables are socioeconomic indicators measuring consumption per capita, unemployment, and informality. The policy variable of interest is tariff reduction of a district and its neighboring areas. The weighted tariff is constructed by combining the information of the number of workers in the district involved in economic activity relative to the total number of workers in the district, and the tariff linked to the economic activity. Controlling for education, migration, rural-urban differences, and initial conditions of socioeconomic development and tariff levels, we find a positive association between tariff reduction and consumption per capita, while the results for unemployment and informality are sensitive to empirical specifications. A 5.4% tariff reduction (equals to a sample standard deviation of weighted tariff changes in Peru from 2004-2015) is associated with 0.112% increase in consumption per capita, of which more than half is a result of regional spillovers of neighboring-districts' tariff changes. A (standard deviation) drop in tariffs is associated with 0.024% increase in the unemployment rate, the effects that are mostly due to direct district's tariff changes. The same tariff reduction is related to a decline in informality by 0.314%. Comparing the rural-area sample and the urban-area sample, we find the positive association

1 We thank Elin BaldΓ‘rrago and Hien Nguyen for their help with data. 3 between tariff reduction and consumption (and to a lesser degree for unemployment) to be smaller for the rural areas, while the negative association between tariff reduction and informality to be more significant, suggesting a mixed policy outcomes on trade and inequality in the case of Peru.

JEL Classification Numbers: F1,F13,F14,F16,J2,J46,O17,O24

Keywords: trade, inequality, workers, tariffs

Authors’ E-Mail Addresses: [email protected]; [email protected]; [email protected] 4

Contents Page

ABSTRACT ...... 2

I. INTRODUCTION ...... 5

II. EMPIRICAL SPECIFICATION ...... 7 A. Models ...... 8 B. Estimation procedure ...... 10 C. Instruments ...... 10 D. Spillover effects ...... 10

III. DATA ANALYSIS ...... 11

IV. DISCUSSION ...... 22 A. Economic Significance ...... 22 B. Informality ...... 23 C. Case Studies ...... 23

V. CONCLUSIONS AND POLICY IMPLICATIONS ...... 24

5

I. INTRODUCTION

We integrate three research areas, using economic geography and spatial estimation to analyse trade exposure and neighbourhood effects between districts within the country, and providing empirical evidence on the impact of tariff reduction on socioeconomic indicators. The impact of trade reform on the distributional outcomes, notably consumption, unemployment, and informality have been the focus of much recent research and discussion. (Topalova (2010); Autor, Dorn, and Hanson (2013); Dix-Caneiro and Kovak (2017); McCaig and Pavcnik (2018)). This study uses a case trade reform of Peru during the 2000s to understand further the distributional effects of international trade in developing countries. The empirical analysis uses a combination of socioeconomic indicators, tariff data, and geographic information to measure the spatial interaction of trade-induced economic adjustment across localities. The findings provide several insights on the direct and indirect effects of trade policies on economic activities and contribute to the discussion on the socioeconomic impacts of trade policy for the developing countries.

The critical empirical innovation is to study how levels of socioeconomics in one district are affected by levels in neighbouring areas as well as tariff adjustment; that is, we simultaneously examine the direct and indirect, spatial effects of tariff reduction at the district level. Based on the sample from 2004-2015, the analysis allows for variation in the distributional consequences of international trade (both trade costs and tariff barriers) across districts within Peru dependent on their production activities, geographic location, market access, and neighbouring influences. Our measures of socioeconomics (the dependent variables) are four of the indicators frequently discussed in the policy formulation: income per capita, consumption per capita, unemployment rate, and informality rate.

Spillovers of socioeconomic development between districts are estimated by using a spatially lagged dependent variable as a determinant. In our specification, the spatial lag is a weighted average of socioeconomic levels in neighbouring districts. Thus, socioeconomic development in each district is influenced by that of its neighbours: the spatial lag is endogenous. Also, it is possible that exogenous shocks affecting in one area have effects that spill over to other districts: this association is captured in a spatially correlated error term. Following Kelejian, Murrell, and Shepotylo (2013) we estimate the relationship by generalised spatial two-stage least squares (GS2SLS): an instrumental variable model that accounts the endogeneity of the spatially lagged socioeconomic development and for the spatially correlated 6 shocks. The instruments are the predetermined variables in our specification (geographic location, infrastructure, and market access) and their spatial lags.

Our main findings suggest that (i) district-specific weighted tariff reduction increases consumption per capita, unemployment rate, and lowers informality rate; and (ii) the level of socioeconomic development in a district's neighbours has a significant effect on the district's socioeconomic outcomes. Because any omitted variable affecting socioeconomic indicators may also be spatially correlated (i.e. a missing variable is spatially correlated), the spatial lag of the socioeconomic (dependent) variable may partially represent the missing variable. To address the issue, we allow for the rural-urban distinction across districts and add spatial lags of determinants (district-specific weighted tariffs) - the spatial-Durbin specification. We also verify the findings specifying neighbouring strength in terms of the bordering contiguity and the inverse of the distance between districts.

Section 2 reviews related studies on within-country variation in trade exposure that suggest a wide range of the effects of tariff reduction on consumption, unemployment, and informality. Section 3 describes data and reports estimates of trade exposure for the Peruvian districts. Section 4 discusses the economic significance and case studies. Section 5 provides the concluding remarks. Appendix A contains a detailed description of data, computing codes, tables and figures discussed in the paper. Appendix B provides the presentation slides for overviewing our study.

Figure 1. Exports and Imports of Peru

7

II. EMPIRICAL SPECIFICATION

In this section, we provide a generic model for within-country trade exposure used in recent studies. We then extend the baseline model to allow a given district's socioeconomic level directly affected by district-specific weighted tariff reduction and indirectly affected by 8 neighbouring districts' socioeconomic levels. Since the relationship holds for all districts, it implies via a reduced form that each district's socioeconomic level is related to the tariff reduction of all districts. With these spatial interactions measured, we can test the null hypothesis of no spillovers in socioeconomic development and trade exposure across districts. We also describe our empirical specification for addressing the endogenous spatially lagged dependent variable, based on the generalised spatial two-stage least squares, a spatial form of GMM estimator with instruments drawn from geographic information on location, infrastructure, and market access in Peru.

The literature, notably Topalova (2010), Autor, Dorn, and Hanson (2013), Dix-Caneiro and Kovak (2017), McCaig and Pavcnik (2018), argue that trade exposure encountered by workers has an economic effect on consumption, unemployment, and informality in the labour markets. Generally, a distinction is made on location-specific trade exposure that is subject to the concentration of workers in a given industry. In our analysis, the location-specific trade exposure is captured by industry-specific tariff reduction weighted by the number of workers in the respective industry-location. We also control for fundamentals across locations (districts) using education (human capital accumulation), immigration (worker mobility), rural-urban differences in the estimation.

We study on consumption, unemployment, and informality, excluding income for a reason. As pointed out by Meyer and Sullivan (2017), official income inequality statistics do not necessarily agree with consumption inequality. The income statistics do not accurately reflect inequality because income is poorly measured, particularly in the tails of the distribution, and current income differs from permanent income, failing to capture the consumption paid for through borrowing and dissaving and the consumption of durables. We expect the issues with income statistics to be especially pronounced in the Peruvian data and therefore focus on consumption in the analysis.

A. Models

Our model estimates two spatial processes: (a) spatial spillovers between socioeconomic levels and (b) spillovers between idiosyncratic aspects of trade exposure that affect socioeconomic levels.

= + + +

𝑦𝑦𝑑𝑑 𝑋𝑋𝑑𝑑𝛽𝛽1 𝐻𝐻𝛽𝛽2 πœ†πœ†πœ†πœ†π‘¦π‘¦π‘‘π‘‘ 𝑒𝑒𝑑𝑑 9

= + ; = 1, . . . , where is an Γ— 1 vector of observations𝑒𝑒𝑑𝑑 πœŒπœŒπœŒπœŒπ‘’π‘’ on𝑑𝑑 socioeconomicπœ–πœ–π‘‘π‘‘ 𝑑𝑑 𝑇𝑇 levels for districts. is a Γ— ( ) matrix of observations on fundamentals whose values vary over time; is an 𝑦𝑦𝑑𝑑 𝑛𝑛 𝑛𝑛 𝑋𝑋𝑑𝑑 Γ— ( ) matrix of observations on fundamentals whose values do not vary over time; is 𝑛𝑛 π‘˜π‘˜1 π‘˜π‘˜1 𝐻𝐻 an Γ— weighting matrix; is the Γ— 1 disturbance vector; and is the exogenous shock 𝑛𝑛 π‘˜π‘˜2 π‘˜π‘˜2 π‘Šπ‘Š vector. and are, respectively, Γ— 1 and Γ— 1 parameter vectors. and are scalar 𝑛𝑛 𝑛𝑛 𝑒𝑒𝑑𝑑 𝑛𝑛 πœ–πœ–π‘‘π‘‘ parameters. is the spatial lag of the dependent variable . In this framework, depends 𝛽𝛽1 𝛽𝛽2 π‘˜π‘˜1 π‘˜π‘˜2 πœ†πœ† 𝜌𝜌 on , as does, and thus they are both endogenous. π‘Šπ‘Šπ‘¦π‘¦π‘‘π‘‘ 𝑦𝑦𝑑𝑑 𝑦𝑦𝑑𝑑 𝑑𝑑 𝑑𝑑 𝑒𝑒 π‘Šπ‘Šπ‘¦π‘¦ We define the weighting matrix so that the value of the variable (dependent and control variable of interest) for each district depends directly on an average of that variable for π‘Šπ‘Š bordering districts (and alternatively, subject to an inverse distance to other districts). Suppose the district has districts that border it. Then, the row of has zeroes everywhere exceptπ‘‘π‘‘β„Ž in positions corresponding to neighbours. Inπ‘‘π‘‘β„Ž these positions, the values in the 𝑖𝑖 πœ™πœ™π‘–π‘– 𝑖𝑖 π‘Šπ‘Š π‘‘π‘‘β„Ž row are πœ™πœ™; 𝑖𝑖a weighting matrix is row normaπœ™πœ™π‘–π‘– lised, so the elements of each row sum to unity.𝑖𝑖 The 1 diagonalπœ™πœ™ elements𝑖𝑖 of are zero: no district is considered as its own neighbour.

π‘Šπ‘Š If | | < 1 and | | < 1 and let = ( ) , the solution for is: βˆ’1 πœ†πœ† 𝜌𝜌 =𝐺𝐺 𝐼𝐼 βˆ’+πœ†πœ†πœ†πœ† + 𝑦𝑦𝑑𝑑

𝑦𝑦=𝑑𝑑 ( 𝐺𝐺𝑋𝑋𝑑𝑑𝛽𝛽1) 𝐺𝐺𝐺𝐺;𝛽𝛽2= 1𝐺𝐺, .𝑒𝑒. 𝑑𝑑. , βˆ’1 implying that the value of for𝑒𝑒 each𝑑𝑑 district𝐼𝐼 βˆ’ 𝜌𝜌𝜌𝜌 dependsπœ–πœ–π‘‘π‘‘ 𝑑𝑑 on all the𝑇𝑇 fundamentals (including exposure to tariff reduction) of each district, as well as the shocks in all the districts. If = 0, 𝑦𝑦𝑑𝑑 there are no spillovers in the fundamentals between districts. If elements are i.i.d. over πœ†πœ† districts with ( = 0, = , the variance-covariance matrix of is πœ–πœ–π‘‘π‘‘ 2 π‘šπ‘šπ‘šπ‘šπ‘šπ‘šπ‘šπ‘š 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣= 𝜎𝜎( ) ( ) 𝑒𝑒𝑑𝑑 2 βˆ’1 βˆ’1 such that the elements are spatially𝑉𝑉𝐢𝐢𝑒𝑒 correlated.𝜎𝜎 𝐼𝐼 βˆ’ 𝜌𝜌 𝜌𝜌 𝐼𝐼 βˆ’ πœŒπœŒπœŒπœŒβ€²

10

B. Estimation procedure

An IV procedure is used to address the spatial lag term , which is endogenous. The estimation involves three steps: (1) , , , are estimated by an IV procedure, accounting for π‘Šπ‘Šπ‘¦π‘¦π‘‘π‘‘ the spatial correlation of ; (2) the estimated parameters are used to estimate and ; (3) the 𝛽𝛽1 𝛽𝛽2 πœ†πœ† 𝜌𝜌 estimated is used to transform model, which is then estimated by an IV procedure. 𝑒𝑒 𝑒𝑒 𝜌𝜌 𝜌𝜌 C. Instruments

Using the above,

( ) = + ( ) + ( )+. .. βˆ’1 2 2 It follows that 𝐼𝐼 βˆ’ 𝜌𝜌𝜌𝜌 𝐼𝐼 πœ†πœ†πœ†πœ† πœ†πœ† π‘Šπ‘Š

[ ] = (( ) )[ + ] βˆ’1 , and thus 𝐸𝐸 𝑦𝑦𝑑𝑑 𝐼𝐼 βˆ’ 𝜌𝜌𝜌𝜌 𝑋𝑋𝑑𝑑𝛽𝛽1 𝐻𝐻𝛽𝛽2

[ ] = ( + ( ) + ( )+. . . )[ + ] 2 2 Following Kelejian et𝐸𝐸 al.π‘Šπ‘Š (2013)𝑦𝑦𝑑𝑑 π‘Šπ‘Š, 𝐼𝐼, , πœ†πœ†πœ†πœ†, πœ†πœ† areπ‘Šπ‘Š used as the𝑋𝑋𝑑𝑑 𝛽𝛽instruments.1 𝐻𝐻𝛽𝛽2

𝑑𝑑 𝑑𝑑 𝑋𝑋 𝐻𝐻 π‘Šπ‘Šπ‘‹π‘‹ π‘Šπ‘Šπ‘Šπ‘Š D. Spillover effects

As ( , ) change in a given district , we can calculate direct effects, which in turn are transmitted to other districts through the spatial lag and feedback to the district , resulting in 𝑋𝑋𝑑𝑑 𝐻𝐻 𝑖𝑖 indirect effects. If 0, the level of socioeconomics in every district depend on the 𝑖𝑖 fundamentals of all areas in Peru. πœ†πœ† β‰  Suppose there is a tariff reduction in Arequipa, corresponding to the 1st column

(regressor) of . Denote the tariff reduction at time for Arequipa as , , then the change in the element of with respect to is 𝑋𝑋𝑑𝑑 , 𝑑𝑑 π‘₯π‘₯𝑑𝑑 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 π‘‘π‘‘β„Ž 𝐽𝐽 𝑦𝑦𝑑𝑑 π‘₯π‘₯𝑑𝑑 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝐽𝐽 = , ; = 2, . . . , , 𝑑𝑑 πœ•πœ•π‘¦π‘¦ 1 1 𝐽𝐽1 𝑑𝑑 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑏𝑏 𝐺𝐺 𝐽𝐽 𝑁𝑁 , where , is the 1st element ofπœ•πœ•π‘₯π‘₯ and is the ( , 1) element of . These spillovers (emanating effects) are calculated with respect to the meanπ‘‘π‘‘β„Ž conditional on the exogeneous 𝑏𝑏1 1 𝛽𝛽1 𝐺𝐺𝐽𝐽1 𝐽𝐽 𝐺𝐺 variables and is not involved (that is, unrelated to the error terms). 𝑦𝑦𝑑𝑑 𝜌𝜌 11

III. DATA ANALYSIS

We evaluate the effects of tariff reduction on socioeconomic indicators across from 2004-2015. Tariff reduction is a policy intervention of interest as it is associated with higher import competition and trade exposure of workers subject to industries and district locations. The sample is comprised of matched trade and tariff data with socioeconomic indicators at the district level. We source the information to construct our sample as follows.

(1) Geographic information of Peruvian districts is from UNOCHA. (2) Weighted average tariffs are calculated using the import shares and MFN tariffs recorded in the 10-digit product classification (HS 2007) from WITS. (3) Following Dix-Carneiro and Kovak (2017) and Topalova (2010), we match tariff data to district-level economic activities using 4-digit ISIC Rev 4 recorded in the household surveys. The matching is based on the concordance table of the UN Statistics Division. (4) Socioeconomic indicators are from the National Household Surveys (Encuestas Nacionales de Hogares; ENAHOs). Our baseline indicators include consumption per capita, unemployment rate, and informality rate. From the household observations, we calculate district-specific averages and interpolate missing values from the districts' corresponding provincial averages.

We study these indicators as they are most emphasised in the discussions of trade exposure and economic inequality. In the baseline results, we report the association between tariff reduction ( ), district fundamentals ( ), and economic indicators ( ). We then undertake robustness checks on the spatial elements of tariff reduction ( ) and socioeconomic 𝑋𝑋𝑑𝑑 𝐻𝐻 𝑦𝑦𝑑𝑑 indicators ( ) accounting for district geography and spillover effects using the spatial π‘Šπ‘Šπ‘‹π‘‹π‘‘π‘‘ estimation. π‘Šπ‘Šπ‘¦π‘¦π‘‘π‘‘

A. A. Data

A.1 Geography The geographic data is in the system of UBIGEO (CΓ³digo UbicacΓ­on GeogrΓ‘fica), which is a geographic location coding system in Peru used by National Statistics and Computing Institute (Instituto Nacional de EstadΓ­stica e InformΓ‘tica INEI). There are three levels of administration: departments (25), provinces (196), and districts. We use the information of 12 districts based on UNOCHA: 1,873 districts (Sistema Nacional de EstadΓ­stica: 1,869 districts). With the shapefile, we organise the district-level data using UBIGEO. For instance, one of the largest coppers and zinc mine, the Antamina mine, is in San Marcos district (UBIGEO: 021014), Huari (UBIGEO: 0210) province, Ancash (UBIGEO: 02) department (region).

We use 8 geographic data to further account for geography of economic development: mining sites (mineros), gas and oil pipelines (gasoducto, oleoducto), roads (via), airports (aeropuertos), ports (puertos), navigable rivers (rio navegables), and high elevation lines (de lineas de elevacion (curvas de nivel)). The exposure of workers to trade and tariff changes is subject to their economic activities, which, in turn, is influenced by geographic location and access to other regions. For instance, given that mining is the most essential sector of Peru's exports, we expect that the association between tariff reduction and income is conditional on a district's connection to the mining activities.

A.2 District-Specific Weighted Tariffs

13

Table 1. Tariff Changes across Industries in Peru

2004 2015 Animals 13.1 1.5 -11.6 Vegetables𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 12.7 1.9 -10.8π›₯π›₯ Food products 13.7 2.0 -11.7 Minerals 8.5 .8 -7.7 Fuels 6.4 1.2 -5.2 Chemicals 7.2 1.6 -5.6 Plastics & rubbers 11.1 2.8 -8.2 Hides & skins 9.3 2.4 -6.9 Wood 16.0 6.3 -9.7 Textiles & clothing 15.6 5.6 -9.9 Footwear 8.3 1.5 -6.8 Stone & glass 7.8 .8 -7.0 Metals 7.4 .6 -6.8 Machinery 8.1 1.0 -7.0 Transportation 9.8 1.7 -8.2

Table 1 provides an overview of tariff changes in Peru. Across sectors, tariffs dropped from 6.4%-16.0% in 2004 to .6%-6.3% in 2015, amounting to 5.2%-11.7% tariff reduction. To measure the exposure of district workers to the change in tariffs, we construct a weighted tariff at the district level:

, ; ; ; = ; βˆ‘π‘–π‘– 𝐿𝐿𝑑𝑑 𝑖𝑖 2004 πœπœπ‘–π‘– 𝑑𝑑 where is the number of workers𝑋𝑋𝑑𝑑 𝑑𝑑 in district involved in economic activity in the year , ; 𝑇𝑇𝐿𝐿𝑑𝑑 2004 2004; is the total number of workers in district in 2004, and is the tariff linked to 𝐿𝐿𝑑𝑑 𝑖𝑖;2004 𝑑𝑑 ; 𝑖𝑖 the economic activity in year . 𝑇𝑇𝐿𝐿𝑑𝑑 2004 𝑑𝑑 πœπœπ‘–π‘– 𝑑𝑑

𝑖𝑖 𝑑𝑑 For an industry that has no tariff line, we assign it non-tradable and apply a zero tariff accordingly. A more subtle issue is the worker share, , ; . We fix the composition of workers 𝐿𝐿𝑑𝑑 𝑖𝑖;2004 in each district in the initial year 2004. As such, we abstract𝑇𝑇𝐿𝐿𝑑𝑑 2004 from the effect of tariff reduction on the adjustment of worker share across economic activities. Consider an import-competing industry: a significant tariff reduction could reduce the worker share in this industry over the years, thereby inducing the endogeneity of the weighted tariffs across localities.

14

15

A.3 Socioeconomic Data

Our socioeconomic indicators of interest include consumption per capita, unemployment, and informality. The objective is to examine whether there is an association between these variables and tariff adjustment. We include education, immigration, and rural- urban differences as district-specific controls. The data are drawn from the household survey from 2004 to 2015. The panel sample is not balanced due to missing observations. For our analysis, we interpolate the missing district-level observations with the provincial-level averages; this procedure results in a sample of 1,873 districts with all the variables entirely available for the estimation.

Note: Of 1,873 districts, 721 have data for all socioeconomic indicators from the household surveys. We interpolated for missing data the district-level observations using the averages of their respective provinces. While this is a second-best practice, it allows construction of full spatial distance/contiguity matrix to study regional spillovers.

16

B. Estimation of Tariff Reduction: Direct Effects and Neighboring-District Spillovers

To examine the association between tariff changes and socioeconomic indicators, we consider both direct and indirect linkages. The exposure of workers is varying across the district due to product-specific tariff reduction and location-specific economic activities. Workers in districts negatively affected from tariff reduction, and thus higher import competition, would likely migrate to the less-affected areas, thereby contributing to the spillovers of tariff reduction; this happens even though workers in the less-affected districts were not exposed to the tariff reduction directly.

As such, it is useful to include the geography of economic activities in the analysis to account for the relocation and re-allocation of factors, workers and productive capitals following the tariff changes. Empirically, we require a proxy to quantify the movement of these factors. 17

The migration of workers across geographic regions may be discouraged by the government's garden-variety policies and regulations, of which the implementation and effectiveness of could differ across time and countries. We address this empirical challenge by estimating the spatial effects across districts.

Note: Regional spillovers/spatial association depend on the spatial weights used in the analysis. We can construct the spatial weights using contiguity approach and distance approach. The choice may be data-driven and context specific. The estimation issues may be further complicated by geographic attributes, i.e. points (mining sites, airports) vis-a-vis lines (road networks and rivers). 18

19

B.1 Baseline Estimation

In the baseline estimation, we first test whether OLS is enough. Formally, this is done by the Moran test for determining whether the residuals of a model fit by OLS are correlated with neighbouring residuals. We create the spatial weighting matrices; contiguity matrix (nearby = shares a border) as well as distance matrix . Based on the contiguity weights, we calculate the test statistics whether to accept or reject that the residuals from the model above are . . . π‘Šπ‘Š 𝑀𝑀 ( ; the null hypothesis). 𝑖𝑖 𝑖𝑖 𝑑𝑑

𝐻𝐻0 If the test rejects that the residuals from the model above are . . ; the residuals are correlated with neighbouring residuals as defined by the weights, we fit the spatial models in 𝑖𝑖 𝑖𝑖 𝑑𝑑 which the observations are not independent as defined by the weighting matrix, thereby allowing for spillovers from neighbouring districts. There are several estimation choices of spatial models. The generalised spatial two-stage least squares (gs2sls) model is robust to violations of normality; the maximum likelihood (ml) model is more efficient when the errors are normally distributed. We consider two variants:

β€’ a spatial lag of independent variables β€’ a spatially autoregressive error

Based on the estimation, we assess the average effects, for instance, denoting a, b, c from the recursive process: On average, for a 1-percentage-point change in the weighted tariff: the own-district direct effect is to change the dependent variable by a percentage point; the indirect spillover effect reduces the dependent variable by b percentage points. The cumulative effect is c percentage points.

Table 2. OLS Estimation (1,873 districts)

log(Consumption) Unemployment Informality rural district -.073 (.011) -.030 (.005) .037 (.018) 2 migration rate π‘Œπ‘Œ 1.316 (.171) .539 (.074) .305 (.269) log(education) .585 (.044) .233 (.019) -.081 (.069) -2.770 (.211) -.690 (.091) 1.086 (.331) : -1.905 (.146) -.469 (.063) .718 (.229) 0 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑓𝑓 .435 .332 .020 0 2 Additionalπ›₯π›₯ models:π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯2 𝑓𝑓 (ii) Incl. initial 𝑅𝑅dependent ( ) .335 (.022) .336 (.021) .070 (.020) π‘Œπ‘Œ0 π‘Œπ‘Œ0 20

: -1.965 (.138) -.251 (.061) .674 (.229) (iii) Incl. ( ) and rural inter. 0 2 π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯𝑓𝑓 : -1.409 (.228) .180 (.097) -.043 (.375) 0 π‘Œπ‘ŒΓ— -.604 (.197) -.481 (.085) .782 (.324) 0 2 (iv) Growthπ›₯π›₯ π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯as a dependent𝑓𝑓 ( ) π‘Ÿπ‘Ÿπ‘Ÿπ‘Ÿπ‘Ÿπ‘Ÿπ‘Ÿπ‘Ÿπ‘Ÿπ‘Ÿ: ) -.971 (0.121) .039 (.058) .415 (.264) 0 Standard errors in parentheses. Dependentπ‘Œπ‘ŒΜ‡ variables are district-level, averaged over 2012-16 (t=2). π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯𝑓𝑓0 2 Explanatory variables (including tariff: initial) are district-level, calculated as the 2004-07 (t=0) averages. Tariff changes (weighted) are district-level measured by 2012-16 (t=2) average minus 2004-07 (t=0) average.

Table 2 reports the OLS estimation using the three socioeconomic indicators separately as dependent variables. In the OLS estimation, we exclude the spatial lag of the dependent variable ( ). The coefficients of tariff reduction : are significant statistically: tariff reduction increases consumption per capita and unemployment and decreases informality π‘Šπ‘Šπ‘¦π‘¦π‘‘π‘‘ βˆ’π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯𝑓𝑓0 2 in the Peruvian labour markets. While the results for unemployment and informality are sensitive to the inclusion of initial dependent variable, interaction with rural variable, or using the growth of the dependent variable, the results that tariff reduction increases consumption are consistent across the OLS specifications.

Table 3. 2SLS Estimation (1,873 districts)

Y log(Consumption) Unemployment Informality initial dep. .334 (.022) .296 (.031) .052 ( .022) rural district -.055 (.015) -.007 (.006) -.029 ( .026) 0 migration rateπ‘Œπ‘Œ 1.065 (.201) .345 (.086) 1.022 ( .354) log(education) .177 (.068) .100 (.023) .197 ( .105) -2.210 (.768) -1.048 (.384) 5.717 (1.384) Λ† -1.873 (.589) -.756 (.292) 4.270 (1.055) 0: 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑓𝑓 .496 .391 .013 0 2 coeffsπ›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯ of 2 𝑓𝑓 Λ† 𝑅𝑅 : for rural sample -4.818 (1.755) -.685 (.598) 7.987 (2.936) 0 2 forπ›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯ urban𝑓𝑓 sample -2.198 ( .794) -.422 (.439) 3.239 (1.365) Λ† ruralΓ— : .790 ( .440) .288 (.185) 1.504 (0.792) Standard errors in parentheses. Dependent variables are district-level, averaged over 2012-16 (t=2). π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯𝑓𝑓0 2 Explanatory variables (including tariff: initial) are district-level, calculated as the 2004-07 (t=0) averages. Tariff changes (weighted) are district-level measured by 2012-16 (t=2) average minus 2004-07 (t=0) average. Instruments are geographic variables, including distances to mining sites, ports, and airports.

21

Table 3 reports the 2SLS estimation. We use as instruments the geographic variables, including district-specific distances to mining sites, ports, and airports. In the 2SLS estimation, we again exclude the spatial lag of the dependent variable ( ). The coefficients of tariff reduction are significant statistically: tariff reduction increases consumption per : π‘Šπ‘Šπ‘¦π‘¦π‘‘π‘‘ capita and unemployment and decreases informality in the Peruvian labour markets. The 2SLS βˆ’π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯𝑓𝑓0 2 results for consumption, unemployment, and informality are insensitive to the inclusion of initial dependent variable and interaction with a rural variable. In particular, the results that tariff reduction increases consumption and reduces informality are more pronounced in the sample of rural Peruvian districts.

B.2 Spatial Estimation and Geography of Economic Activities

As the diagnostic tests suggest spatial dependence in the data, we now examine more robust estimation to account for the regional spillovers. Also, it is useful to consider a spatial lag model that, besides the spatial lag of the dependent variable, also includes other non-spatial endogenous regressors. We posit that the weighted tariff is endogenous and use geographic information captured by district-specific distance to the mining sites, ports, and airports as the instruments.

Table 4. Spatial 2SLS Estimation (1,873 districts)

Y log(Consumption) Unemployment Informality initial dep. .323 (.025) .306 (.038) .054 (.023) rural district -.057 (.015) -.010 (.006) -.024 (.023) 0 migration rateπ‘Œπ‘Œ 1.046 (.188) .334 (.085) 1.015 (.328) log(education) .197 (.071) .111 (.022) .171 (.099) -1.998 (.792) -.716 (.380) 5.428 (1.233) Λ† . -1.718 (.607) -.504 (.287) 4.041 ( .942) 0 neighbors𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑓𝑓 Y .183 (.089) .199 (.074) -.268 ( .355) 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 β„Žπ‘”π‘” .498 .407 .014 Standard errors in2 parentheses. Dependent variables are district-level, averaged over 2012-16 (t=2). 𝑅𝑅 Spatial weights measured as the inverse distance between districts. Instruments: mining sites, ports, airports. Explanatory variables (including tariff: initial) are district-level, calculated as the 2004-07 (t=0) averages. Tariff changes (weighted) are district-level measured by 2012-16 (t=2) average minus 2004-07 (t=0) average.

Table 4 reports the spatial 2SLS estimation. We again use as instruments the geographic variables, including district-specific distances to mining sites, ports, and airports. In the spatial 2SLS estimation, we include the spatial lag of the dependent variable ( ). The coefficients of

π‘Šπ‘Šπ‘¦π‘¦π‘‘π‘‘ 22

tariff reduction : are significant statistically: tariff reduction increases consumption per capita and decreases informality in the Peruvian labour markets. The 2SLS results for βˆ’π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯π›₯𝑓𝑓0 2 unemployment are no longer statistically significant. Further, we also find that the coefficients of the spatially lagged dependent variables ( ) are statistically significant for consumption per capita and unemployment. To understand the spatial effect, focus on the consumption π‘Šπ‘Šπ‘¦π‘¦π‘‘π‘‘ equation. Suppose that districts 1 and 2 are similar except that district 1 is surrounded by districts at high levels of consumption per capita while district 2 is surrounded by districts at low levels of consumption per capita. The effect of having high-consumption neighbouring districts rather than low-consumption neighbouring districts increases the consumption per capita of district 1 relative to district 2. We still find that the coefficient of tariff reduction on consumption per capita is much more significant than the neighbouring spillovers.

IV. DISCUSSION

A. Economic Significance

In terms of statistical robustness, we find from the OLS, the 2SLS, and the spatial 2SLS estimation that tariff reduction increases consumption per capita, while the results for unemployment and informality are sensitive to the empirical specifications. Given the estimates, we examine the 'Direct' and 'Spillover' effect coefficients and, as shown in Table 5, find that a 5.4% tariff reduction (equals to a sample standard-deviation of weighted tariff changes in Peru from 2004-2015) is associated with 0.112% increase in consumption per capita, of which more than half is a result of regional spillovers of neighbouring-districts' tariff changes. We also find that a (standard deviation) drop in tariffs is associated with 0.024% increase in the unemployment rate, the effects that are mostly due to direct own-district's tariff changes. The same tariff reduction would reduce informality by 0.314%.

The estimates suggest that the tariff reduction of Peru is favourable to the overall consumption per capita across districts. Unlike in some other countries, this latest episode of tariff reduction in Peru progressed in isolation from other structural uncertainties and macroeconomic reforms. Thus, it is more likely that the positive effects of import competition were realised in the form of higher consumption for the period of this study. While the positive findings on consumption could be unique to Peru, we find that more import competition in the form of lower tariffs is associated with a higher unemployment rate in the estimation. Our findings need further supportive evidence on the aggregate welfare effects, 23 accounting for costs of living, labour market adjustment, relative factor incomes, and resource reallocation. Also, beyond the scope of this exercise are potential effects of tariff reductions through - (i) productivity adjustment and access to foreign markets; - (ii) markup variation of domestic and international producers; - (iii) differential adjustment speed of socioeconomic indicators (i.e. income vis-a-vis employment). Table 5. Effects of tariff reduction on Peruvian districts, 2004-2015 socio indicators Direct Spillover Total = 5.431 consumption (log) +1.718 +.360 +2.079 +.112% βˆ’π›₯π›₯1𝑠𝑠𝑠𝑠𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 βˆ’ 𝑝𝑝𝑝𝑝 unemployment +.504 -.046 +.457 +.024% informality -4.041 -1.754 -5.796 -.314%

B. Informality

For Peru, about of working population (ILO) and of registered firms competing with 2 3 informal firms (World Bank3 Enterprise Surveys). These statistics4 are not uncommon in developing countries. The implementation of e-payroll in August 2007 increased the number of registered employment significantly. Faster growth of the labour-intensive industries, rigid labour laws, and complex tax and small business regulations discourage the formal sector from growing (The Economist, 2018).

The size of the informal sector is subject to not only the measurement error but also the time-varying nature of migration. Take Gamarra (La Victoria district of ; UBIGEO: 150115) as an example: some anecdotal evidence suggests that immigrants from Venezuela have recently driven up the activities of the informal sector there, notwithstanding the efforts of the Peruvian government to formalise the small businesses. The government also attempts to increase the pool of skilled workers by providing tax incentives for worker training, which also encourages informal firms to register.

C. Case Studies

C.1 Gamarra

Located in La Victoria district (UBIGEO 150115) of Lima, Gamarra is the largest market of fabrics and clothes and a host of informal small businesses in Peru. More than 15,000 24 businesses, Gamarra exports textiles to LATAM and global brands but has recently been overwhelmed by highly competitive Chinese textiles imports. The Peruvian cotton is high- quality but expensive: its production fell, growers moved to other profitable crops. Textiles account for 3.0% (9.3%) of exports in 2017 (2003) and of 4.9% (5.0%) imports in 2017 (2003).

C.2 Maca

Farmed in JunΓ­n, Maca, a pungent, turnip‑like vegetable and heralded as a cancer-fighting and libido-boosting food, has become in high demand in the US and Chinese markets. High demand in China has encouraged cases of smuggling out of Peru illegally. On shelves of US supermarkets (Whole Foods) during 2012-13, supplier prices of Maca went up from USD3.6 to 20/pound. Maca workers pay went up from USD9.65/day to only 11.37/day.

C.3 Laredo

There are two types of agricultural districts in Peru: one in the agro-exporting coast and another with less agro-exporting in the highlands. An example of the first is the asparagus producing Laredo (UBIGEO 130106) and of the second the cattle-raiser Γ‘uΓ±oa (UBIGEO 210806). In Trujillo province, Laredo is an agro-exporting coastal district. It competes with Mexican asparagus for the US markets. The district is the location of agro-industrial Sol de Laredo produces sugar & derivatives. Historically, migrants formed small-scale manufacturers and helped establish the nearby Trujillo city as the centre of the Peruvian footwear industry.

C.4 La Rinconada

La Rinconada, in Ananea district (UBIGEO 211002) of San Antonio de Putina province, is a town near a gold mine. In the 2000s, its population increased by more than 200% driven by the global price of gold. Miners are paid on the 31st day on however much ore they can carry. Surrounding areas are intoxicated with cyanide, mercury and sewage

V. CONCLUSIONS AND POLICY IMPLICATIONS

This study contributes to the literature by examining the unique experience of Peru in the 2000s, the period that was characterised by the non-existence of other simultaneous reforms. Using our detailed data and analysis of tariff reduction across industries, together with an 25 account of neighbouring spillover effects across districts, we find that the tariff reduction in Peru raised consumption, while lowered informality and thus increased unemployment.

The effects of tariff reduction on socioeconomic outcomes could involve other complexities not addressed in this study. Likely,affluent districts are likely more exposed to tariff reductions since the most impoverished regions are rural, isolated, and are more involved in self-sufficient agriculture. The rural districts do not produce for the urban markets, so they are less affected by reductions in tariffs of competing industries that pass through to national prices, while benefiting from cheaper imported consumer and investment goods. The agricultural sector benefited the most from lower tariffs to Peruvian imports from the United States (and from China) and thus may explain why lower tariffs are related to higher consumption. This linkage deserves further examination.

Further, there are several channels that tariffs may affect consumption. Tariff reduction was most significant in agricultural districts, and agricultural exporting was among the most dynamic sectors in Peru. The relationships between tariff reduction and expenditures depend on whether competition (producer's markup) increases (declines) as a result. Competition lowers prices and increases real incomes. This issue may be complicated by the labour mobility and the changing consumption baskets (tradables v. others) of the poor households relative to the other groups.

It may also be the case that agricultural and textile producers are exposed to lower import tariffs for their products to the domestic market (usually of low quality) and switched to export products and more profitable items. For example, it may be the case that rice producers, which are entirely unproductive in the coast of Peru due to scarce rain, faced with import competition were forced to move to asparagus and that set them into an altogether more profitable path.

Equally intriguing are the findings on unemployment and informality. The oft-cited argument is the informal sector is a labour-market buffer for unemployment. That informality declined while unemployment increased with the tariff reduction is consistent with this argument. Due to frictions in the labour markets, workers in sectors negatively affected by import competition may not be readily moved and re-allocated to other industries in the economy. Without the buffer services, the Peruvian workers are exposed to import competition. However, reducing informality entails worker benefits, notable improvement in labour standards, product quality, business registry, and tax collection. 26

Overall, the analysis highlights the role of geography, market access, and infrastructure. Recently, Peru has received significant foreign investment in hydrocarbons and infrastructure projects. This evidence implies that the neighbouring spillovers found in this study may become even more critical to the susceptibility of workers across the Peruvian districts to international trade. Finally, this study largely abstracts from the possible role of other macroeconomic factors. For instance, a currency adjustment can undo much of the tariff effects identified; an appreciation of the sol would encourage imports while suppressing the profits of local producers and exporters. While Peru is a success story of Latin America of the last decade, underlined by countercyclical and prudent macro policies, its dependence of commodity exports for growth, widespread inequality and poverty exemplified by small middle-class consumers and SMEs, may yet expose the economy to external risk factors in the foreseeable future. Such external exposure will continue to characterise not only Peru but also most of developing economies. Bridging microdata with macro factors thus deserves further studies and better understanding.

27

References

1. Arkolakis, Costas, Costinot, Arnaud, Donaldson, Dave, and Rodriguez-Clare, Andres, 2019, The Elusive Pro-Competitive Effects of Trade, Review of Economic Studies, 86, 46- 80.

2. Autor, David, Dorn, David, and Hanson, Gordon, 2013, The China Syndrome: Local Labor Market Effects of Import Competition in the United States, American Economic Review, 103, 6, 2121-68.

3. BaldΓ‘rrago, Elin, and Salinas, Gonzalo, 2017, Trade Liberalization in Peru: Adjustment Costs Amidst High Labor Mobility, IMF Working Paper 17/47.

4. Brülhart, Marius, Carrère, Céline, and Trionfetti, Federico, 2012, How wages and employment adjust to trade liberalisation, Journal of International Economics, 86, 68-81.

5. Dell, Melissa, 2010, The Persistent Effects of Peru's Mining MITA, Econometrica, 78, 6, 1863-1903.

6. Dix-Carneiro, Rafael, and Kovak, Brian K., 2017, Trade Liberalization and Regional Dynamics, American Economic Review, 107, 10, 2908-2946.

7. Fajgelbaum, Pablo D., and Khandelwal, Amit K., 2016, Measuring the Unequal Gains from Trade, Quarterly Journal of Economics, 131, 3, 1113-1180.

8. Feler, Leo, and Senses, Mine Z., 2017, Trade Shocks and the Provision of Local Public Goods, American Economic Journal: Economic Policy, 9, 4, 101-143.

9. Goldberg, Pinelopi Koujianou, and Pavcnik, Nina, 2007, Distributional Effects of Globalization in Developing Countries, Journal of Economic Literature, 45, 1, 39-82.

10. Goldsmith-Pinkham, Paul, Sorkin, Isaac, and Swift, Henry, 2018, Bartik Instruments: What, When, Why, and How, NBER Working Paper No. 24408.

11. Hausmann, Ricardo, and Klinger, Bailey, 2008, Growth Diagnostics in Peru, CID Working Paper No. 181, Harvard University.

12. Henderson, J. Vernon, Squires, Tim, Storeygard, Adam, and Weil, David, 2017, The Global Distribution of Economic Activity: Nature, History, and the Role of Trade, Quarterly Journal of Economics, 1-50. 28

13. ILO, 2014, Trends in informal employment in Peru: 2004 – 2012.

14. International Monetary Fund, 2018, Country Report No. 18/225, Peru: 2018 Article IV Consultation.

15. International Monetary Fund, 2018, Poverty and Inequality in Latin America: Gains during the Commodity Boom but an Uncertain Outlook (Chapter 5), World Economic Outlook, April.

16. Kelejian, Harry H., Murrell, Peter, and Shepotylo, Oleksandr, 2013, Spatial Spillovers in the Development of Institutions, Journal of Development Economics, 101, 297-315.

17. Kovak, Brian, 2013, Regional Effects of Trade Reform: What is the Correct Measure of Liberalization? American Economic Review, 103, 5, 1960-76.

18. Krugman, Paul, 2009, The Increasing Returns Revolution in Trade and Geography, American Economic Review, 99, 3, 561-571.

19. McCaig, Brian, and Pavcnik, Nina, 2018, Export Markets and Labor Allocation in a Low- Income Country, American Economic Review, 108, 7, 1899-1941.

20. Meyer, Bruce D., James X. Sullivan, 2017, Consumption and Income Inequality in the U.S. Since the 1960s, NBER Working Paper No. 23655.

21. Obstfeld, Maurice, 2016, Get on Track with Trade, Finance & Development, December, 12-16.

22. Partridge, Mark D., Rickman, Dan S., Olfert, M. Rose, and Tan, Ying, 2017, International trade and local labour markets: Do foreign and domestic shocks affect regions differently? Journal of Economic Geography, 17, 375-409.

23. Paz, Lourenço S., 2014, The impacts of trade liberalisation on informal labour markets: A theoretical and empirical evaluation of the Brazilian case, Journal of International Economics, 92, 330-348.

24. PySAL: A Python Library of Spatial Analytical Methods

25. The Economist, 2018, Why it’s hard to reduce informality in Latin America’s labour market: Solving a problem that holds back growth and productivity, February 15. 29

26. Topalova, Petia, 2010, Factor Immobility and Regional Impacts of Trade Liberalization: Evidence on Poverty from India, American Economic Journal: Applied Economics, 2, 1-41.

27. Ulyssea, Gabriel, 2018, Firms, Informality, and Development: Theory and Evidence from Brazil, American Economic Review, 108, 8, 2015-2047.

28. World Bank Enterprise Surveys, 2017, Peru 2017 Country Profile, Washington, D.C.

29. WITS: World Integrated Trade Solution

30. World Trade Organization, 2014, Peru Trade Policy Review, Geneva, Switzerland.