Bank Of Greece Sa · ATHEX:TELL
Bank Of Greece Sa — Technical Presentation (September 2022)
49 slides from Bank Of Greece Sa's technical presentation, Sep 30, 2022
Slide 1: Making subsidies work: rules vs. discretion
Making subsidies work rules vs. discretion Federico Cingano Filippo Palomba Paolo Pinotti Enrico Rettore (Bank of Italy) (Princeton University) (Bocconi University) (University of Padua) Sixth Annual Workshop: ESCB Research Cluster 2 29-30 September 2022
Slide 2: Introduction
Introduction • public subsidies to private firms are a major tool of industrial policy in most countries • $61 bil/year in US, €46.5 bil/year in EU in 2018 • crucial for economic recovery after COVID-19 • Aim: create employment in disadvataged areas • Their effect(ivenes)s is highly debated and diff
Slide 3: What we do
What we do • Study major program of public subsidies to Italian firms (Law 488) • RDD Rationing of funds, applications ranked by a quantitative score • Score summarizes objective criteria + discretional priorities by local politicians • Estimate the effect on marginal firms at the cutoff • Extrapo
Slide 4: Institutional background
Institutional background • Law 488/92: main instrument of industrial policy in Italy, 1996-2007 • policy tool: investment subsidies to firms • €26 billions (constant 2010 prices) over 35 calls for projects • subsidies paid to winning applicant firms in 3 yearly installments • Allocation mechanism •
Slide 5: Institutional background
Institutional background • Applications ranked within each sector-region cell according to a quantitative score aggregating criteria clearly defined ex-ante • 1996-1997 1. Skin in the game: own funds relative to amount requested 2. Job creation: number of jobs created 3. No waste: funds requested re
Slide 6: Data
Data • information on 75k projects from 49k applicant firms (Value: €22 bil.)
Slide 7: Data
Data • information on 75k projects from 49k applicant firms (Value: €22 bil.) • merge with employment data (INPS): 40k projects from 27k firms • merge with balance sheets (CERVED): 33.5k projects from 21.5k firms
Slide 8: Empirical strategy: at the cutoff
Empirical strategy: at the cutoff • Main outcomes of interest: investment, employment, productivity, and firm survival • main threat to identification: firms receiving and not receiving the subsidy may be different • the allocation mechanism generates an ideal RDD • only firms scoring above the cuto
Slide 9: Funds received and effect on investment
Funds received and effect on investment + 39% + €500k
Slide 10: Effect on employment growth
Effect on employment growth • log change in employment at alternative horizons +17% +3% +10% Table
Slide 11: Dynamic effects
Dynamic effects PR gy Log of cumulated investment Cumulated log-change of employment Cumulated log-change of revenues e | . | © | | I ~ | < \ 2 1 . | | | . | ~ | a 1 7 | . | g 14 tig! Ley T t | I ; ° | i r | 1 oy | se a 8 | 2 A 0 1 2 3 4 5 24-12] [-12,0] [0,12] [12,24] [24-36] [36-48] [48-60] [60-72
Slide 12: Empirical strategy: Away from the cutoff
Empirical strategy: Away from the cutoff • Angrist & Rokkanen (JASA 2015) “Wanna get away? Regression discontinuity estimation of exam school effects away from the cutoff” • in RDD, selection is captured by the running variable (𝑠) • match eligible and non-eligible on a set of (“killer”) covariates
Slide 13: Empirical strategy: Away from the cutoff
Empirical strategy: Away from the cutoff • Angrist & Rokkanen (JASA 2015) “Wanna get away? Regression discontinuity estimation of exam school effects away from the cutoff” • in RDD, selection is captured by the running variable (𝑠) • match eligible and non-eligible on a set of (“killer”) covariates
Slide 14: Effects away from the cutoff: Testing
Effects away from the cutoff: Testing • Unconditionally, firm outcomes (here, 𝐿 𝑡,𝑡+6 ) are correlated with S
Slide 15: Effects away from the cutoff: Testing
Effects away from the cutoff: Testing • Conditioning on 𝑥 makes the score ignorable • 𝑥 includes growth predictors (size, age, …) selected manually and with ML Conditional independence
Slide 16: Effects away from the cutoff: Testing
Effects away from the cutoff: Testing • Conditioning on 𝑥 makes the score ignorable, while granting support • 𝑥 includes growth predictors (size, age, …) selected manually and with ML Conditional independence Common support
Slide 17: Effects away from the cutoff: Estimation
Effects away from the cutoff: Estimation • Kline (2011): parametric implementation of matching 𝐸 𝑦 𝑍 = 1 = 𝑥′𝛾1 𝐸 𝑦 𝑍 = 0 = 𝑥′𝛾0 • treatment effect for any 𝒔 ≠ 𝒔ത 𝐸 𝑦1 − 𝑦 0 𝑠 = 𝜃 = 𝛾1 − 𝛾0 ′ 𝐸[𝑥|𝑠 = 𝜃]
Slide 18: Effects away from the cutoff: Results
Effects away from the cutoff: Results • getaway.ado: new Stata package implementing A & R + other extrapolation methods (Palomba, 2022) Potential outcomes Treatment effects
Slide 19: Total policy effects
Total policy effects Recover new jobs/investment combining TE & initial size; compute their cost cost per new job cost per worker-year cost of new investment Cost measure: (thousand €'s) (thousand €'s) (subsidy/investment) CIA set of covariates: manual data-driven manual data-driven manual data-driv
Slide 20: Heterogeneity by class size and age
Heterogeneity by class size and age % Treatment effects New jobs per €100k • Smaller firms generate larger % increases in employment, but larger firms produce more jobs-per-€-of subsidy!
Slide 21: Rules vs. discretion
Rules vs. discretion • heterogeneity by score sub-components 𝑠 𝑟 (rules) and 𝑠 𝑑 (discretion) • Verify the CIA: 𝐸 𝑦 𝑠 𝑟 , 𝑠 𝑑 , 𝑥 = 𝐸[𝑦|𝑥] on both sides of the cutoff Left of the cutoff Right of the cutoff
Slide 22: Rules vs. discretion
Rules vs. discretion • High-on-discretion firms less cost effective than high-on-rules firms at generating new jobs
Slide 23: Rules vs. discretion
Rules vs. discretion • Explanation: local politicians target applicant firms that are smaller and demand larger subsidies - CORRELATIONS
Slide 24: Rules vs. discretion
Rules vs. discretion • Explanation: local politicians target applicant firms that are smaller and demand larger subsidies - LASSO Determinants of Obj score Determinants of political discretion
Slide 25: Rules vs. discretion: Counterfactual policies
Rules vs. discretion: Counterfactual policies • Re-rank applicants under alternative criteria, compute cost-per-job • important assumption: policy invariance (e.g., Heckman, 2010) • consistent with balance in observables before/after introduction of discretion
Slide 26: Rules vs. discretion: Counterfactual policies
Rules vs. discretion: Counterfactual policies • Re-rank applicants under alternative criteria, compute cost-per-job • important assumption: policy invariance (e.g., Heckman, 2010) • consistent with balance in observables before/after introduction of discretion Actual policy cost all regions 179 sout
Slide 27: Rules vs. discretion: Counterfactual policies
Rules vs. discretion: Counterfactual policies • Re-rank applicants under alternative criteria, compute cost-per-job • important assumption: policy invariance (e.g., Heckman, 2010) • consistent with balance in observables before/after introduction of discretion Actual Counterfactual policies policy N
Slide 28: Rules vs. discretion: Counterfactual policies
Rules vs. discretion: Counterfactual policies • Re-rank applicants under alternative criteria, compute cost-per-job • important assumption: policy invariance (e.g., Heckman, 2010) • consistent with balance in observables before/after introduction of discretion Actual Counterfactual policies policy N
Slide 29: Conclusions
Conclusions • Law 488/92 • Positive effect on investment (+39%) and employment (+17%) at the cutoff • Heterogeneity in the effect of subsidies across different types of firms • Large firms more cost-effective than small firms • Rules better than discretion • General lessons • Studyin the heterogenei
Slide 30: Thank you!
Thank you!
Slide 31: Data
Data • information on 75k projects from 49k applicant firms (Value: €22 bil.)
Slide 32: Data
Data • information on 75k projects from 49k applicant firms (Value: €22 bil.)
Slide 33: Data
Data • information on 75k projects from 49k applicant firms (Value: €22 bil.) • merge with employment data (INPS): 40k projects from 27k firms • merge with balance sheets (CERVED): 33.5k projects from 21.5k firms
Slide 34: Empirical strategy: effects at the cutoff
Empirical strategy: effects at the cutoff 1 0 • 𝑦𝑖𝑐 and 𝑦𝑖𝑐 are the potential outcomes of firm i in cell c when scoring above (𝑍𝑖𝑐 = 1) and below the cutoff (𝑍𝑖𝑐 = 0)
Slide 35: Empirical strategy: effects at the cutoff
Empirical strategy: effects at the cutoff 1 0 • 𝑦𝑖𝑐 and 𝑦𝑖𝑐 are the potential outcomes of firm i in cell c when scoring above (𝑍𝑖𝑐 = 1) and below the cutoff (𝑍𝑖𝑐 = 0) • difference in observed outcomes between firms with score 𝑠𝑖𝑐 just above and just below the cutoff 𝑠𝑐ҧ is the…
Slide 36: Empirical strategy: effects at the cutoff
Empirical strategy: effects at the cutoff 1 0 • 𝑦𝑖𝑐 and 𝑦𝑖𝑐 are the potential outcomes of firm i in cell c when scoring above (𝑍𝑖𝑐 = 1) and below the cutoff (𝑍𝑖𝑐 = 0) • difference in observed outcomes between firms with score 𝑠𝑖𝑐 just above and just below the cutoff 𝑠𝑐ҧ is the…
Slide 37: RDD diagnostics: Density tests
RDD diagnostics: Density tests
Slide 38: RDD diagnostics: Balance tests (covariates)
RDD diagnostics: Balance tests (covariates)
Slide 39: RDD diagnostics: Balance tests (sub-scores)
RDD diagnostics: Balance tests (sub-scores)
Slide 40: Parametric estimates
Parametric estimates (1) (2) (3) (4) (5) (6) (7) (8) Specification: linear quadratic Kernel: uniform triangular uniform triangular Group FE: no yes no yes no yes no yes Log-change in employment over 3 years Subsidy 0.088*** 0.104*** 0.101*** 0.104*** 0.120*** 0.107*** 0.114*** 0.105*** (0.019) (0.02
Slide 41: Additional results
Additional results • repeated applicants • susidized firms firms have a lower probability of re-applying overall effect: direct effect + indirect effect (negative) • baseline estimates are lower bound to direct effect • local effects • no significant spillovers on other firms in the same labor mar
Slide 42: Repeated applicants
Repeated applicants at Effect of winning on subsequent participation/winning Total and direct effects on employment wo a ------ co # o------ oo Ny i “J T T T T T tH1 t+2 +3 t+4 +5 1 2 3 4 5 6 @ Prob(Participating) | © Prob(Winning) @ Direct Effect 4 Total Effect
Slide 43: Spillover effects
-.02 Spillover effects Control firms, LLM Other (non-applicant) firms, LLM-Industry -.04 1 @ Corrected © Baseline ®@ Corrected © Baseline
Slide 44: Issues with 488 data
Issues with 488 data • Region-call specific rankings published in the GU mixed two or more «actual rankings», those used in the allocation of funds. • We recovered such rankings exploiting additional information on firm size, operating sector, eligibility for co-financing, and geographical area to c
Slide 45: Applicant characteristics, with/out discretion
Applicant characteristics, with/out discretion
Slide 46: Counterfactual policy effects
Counterfactual policy effects • better targeting of constrained/underdeveloped areas?
Slide 47: Financially constrained firms (in progress)
Financially constrained firms (in progress)
Slide 48: Financially constrained firms (in progress)
Financially constrained firms (in progress) Constrained firms
Slide 49: Financially constrained firms (in progress)
Financially constrained firms (in progress)
Bank Of Greece — Technical Presentation (September 2022) on 6ix