June 29 th, 2017 An offer that you can t refuse? Agrimafias and Migrant Labor on Vineyards in Southern Italy Marica Valente (HU Berlin & DIW Berlin) Stefan Seifert (TU Berlin & DIW Berlin)
The 2011 migration wave in southern Italy Introduction 2
The Mediterranean routes Introduction Source: FRONTEX 3
Caporalato and the 2011 migration wave Introduction Caporalato: system of illegal recruitment and exploitation of underpaid farm labor, run by agrimafias Recruited labor: mostly irregular migrants, vulnerable because illegal and undocumented Arab Spring 2011 migration wave via the Mediterranean: 64,000 migrants (cfr. Sicily: 140,000 foreigners, 14,000 in agriculture) Up to 1/3 refuse identification and leave migrant camps Is there illegal employment on southern Italian vineyards? What are the causal effects of this migrant labor supply shock on employment and wages on vineyards? 4
Introduction How to identify employment of illegal labor? Idea: Estimate causal effects on labor productivity (Y/L) and wages (w) Illegal substitutes legal labor: L, Y = (displacement effect) lllegal added to legal labor: Y = or, L or = (complement effect) Illegal competes with legal labor: w = or (competition effect) If there is employment of illegal labor: Overreported labor productivity Wage dampening or decrease 5
Introduction Results and contribution of this study Results indicate unreported employment and competition between legal and illegal workforce Average causal effects 2011-12: Labor productivity increase of 14% (stat. sign.) Wages 7.3% to 23.8% lower than predicted (not stat. sign.) Contribution: First causal analysis of 2011 migration wave for EU First empirical study on potential employment of illegal labor in farmlands of southern Italy 6
Data Data Unit of interest: Treated South weighted average of Sicily and Apulia Vineyard panel 1991-2012 for 25 regions: 14 Italian + 11 French Vineyards aggregated at regional level (FADN) Outcomes: Labor productivity = Income from crops Total hours worked Hourly wage = Total wage bill Paid hours worked 7
Methodology The causal model Aim: Estimate causal effects of the shock/treatment Y it : outcome (labor productivity or wages) t = 1,..., T 0,, T with T 0 pre-treatment periods i = 1,, J + 1 units, i = 1 is treated in T 0 + 1 Dependent variables (Y): Y 1t TR Treated; Y 1t N Not treated Control variables (X): X 1t,, X J+1,t Treatment effect (Δ): 1t = Y 1t TR Y 1t N, for t > T 0 Problem of causal inference: Y 1t N is not observable for t > T 0 N Estimate counterfactual Y 1t 8
The synthetic control method (SCM, Abadie 2010) Aim: For t > T 0, estimate Y 1t N (outcome had the shock not occurred) Methodology Idea: For t < T 0, build twin (synthetic control) of treated unit w.r.t. Y and X For t > T 0, predict Y 1t N from this twin Estimation: For t < T 0, minimize weighted distance between (Y 1t, X 1t ) and (Y jt, X jt ) Obtain non-negative weights for untreated unit such that σ J+1 j=2 w j = 1 Build Y 1t N (synthetic unit) as weighted linear combination of Y jt using w j SC unit: Y N J+1 1t = j=2 SC estimator: 1t = Y TR 1t w j Y jt Estimated for t > T 0 w j Y jt J+1 j=2 Estimated for t < T 0 9
Methodology Predictors (X) 1. Production function X E.g., land productivity, capital intensity, etc. 2. Labor market X Unemployment rate, unskilled labor share 3. Farm-specific X Paid/family labor 10
Identification assumptions Methodology A1: Exogenous shock (quasi-experimental setting) Southern Italy treated due to geography, no self-selection A2: No spillovers of treatment Fleeing occurs directly after landing A3: No technological progress, no labor market or price shocks Low skill requirements of field picker jobs 11
Labor productivity: Estimated weights for control regions Champagne-Ardenne: 0.007 Veneto: 0.04 Results Emilia-Romagna: 0.171 Marche: 0.108 Abruzzo: 0.318 Languedoc- Roussillon: 0.136 Tuscany: 0.019 Sardinia: 0.2 Own results, Source of Map: Wikipedia 12
Results Labor productivity path: Treated South (solid) vs. synthetic (dotted) ATT 2011,2012 = 0.5 1,2011 + 1,2012 = 14% 13
Placebo inference: Labor productivity gaps Results No large scale (asymptotic) inference small N placebo inference H 0 : ATT j ATT 1 P Value = σ J+1 j=1 1(ATT j ATT 1 ) J = 0. 08 14
Results More inference: ATT over MAD of Treated South and placebos ATT: exceptionally strong given pre-treatment fit (MAD) MAD = 1 T 0 Y T jt Y N jt P Value = σ J+1 j=1 1(ATT j ATT 1 ) 0 J t=1 = 0. 08 s. t MAD j MAD 1 15
Answering our research question Conclusions Labor Productivity Effects Expected: Increase in labor productivity Wage Effects Expected: Decrease in wages Wage dampening X There is causal evidence for the employment of illegal workforce on vineyards after the 2011 migration wave Substitution and/or complement effect of workers Separate analysis for Sicily and Apulia: causal effects stronger in Sicily Little evidence for competition effect that keeps wages close to the minimum level: Wages were already very low (5-6 euros/h) 16
Thank you for your attention. 17
On Methodology: Abadie, A., Diamond, A., Hainmueller, J. (2010). Synthetic control methods for comparative case studies: Estimating the effect of California's tobacco control program. Journal of the American Statistical Association 105, 493-505. Gobillon, L., and Magnac, T. (2016). Regional Policy Evaluation: Interactive Fixed Effects and Synthetic Controls, The Review of Economics and Statistics 98(3), 535-551. On Migration and Illegal Labor: Selected References Peri, G. (2012). The effect of immigration on productivity: Evidence from U.S. states. The Review of Economics and Statistics 94 (1), 348-358. Dustmann, C., Schonberg, U., Stuhler, J. (2016). The impact of immigration: Why do studies reach such different results? Journal of Economic Perspectives 30 (4), 31-56. Tumen, S. (2016). The economic impact of Syrian refugees on host countries: Quasiexperimental evidence from Turkey. American Economic Review 106, 456-60. 18
Appendix 19
Climate, temperatures and precipitations of Italy and France Data http://go.grolier.com/atlas 20
Labor Productivity: Predictor Means Appendix 21
Wage: Predictor Means Appendix 22
Descriptives of predictors (X) Data 23
Estimated weights for control units Results 24