## wpiea2022104-print-pdf

## Source details

**Canonical URL:** [wpiea2022104-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022104-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022104-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022104-print-pdf.pdf.json)

---

### Antitrust decisions, dataset construction, and cartel characteristics
- Role and enforcement (ADLC):
  - Fines set “according to the seriousness of the facts, the extent of the harm done to the economy, the individual situation of the company ... and whether it is an infringement that has been repeated or not”.
  - Fine caps: “10% of the global turnover of the group to which the company that is being fined belongs to” or at a maximum amount capped at 3 million euros if the infringement is committed by an entity other than a for-profit firm.
  - Injunctions: formal notifications to stop anticompetitive behavior.
  - Cross-country cases handled by the Directorate-General for Competition of the European Commission.
- Database construction and scope:
  - Source: ADLC decision PDFs (French); extracted fields: firm names, amount of the fine, type of anticompetitive practice, duration of practice, cause of breakup, year verdict returned, starting year of the investigation.
  - Firms matched to SIREN codes via INSEE/Orbis and Bing-verification; matched to administrative production datasets (FICUS) covering French firms 1994–2007.
  - Exclusions: international cartels (cross-country), cases of abuse of dominant position, repeat offenders, cases notified by injunction alone.
  - Final dataset: 174 cartels and more than a thousand firms (1,037 cartel members reported).
- Cartel prevalence, size, and duration:
  - Average duration ≈ 4.5 years; median ≈ 3 years.
  - Average number of firms per cartel = 6; median = 4; standard deviation = 7; some cartels include > 70 firms.
  - Types of practices (means): Price fixing 0.35; Market allocation 0.29; Production quotas 0.04; Information sharing 0.59; Repeat offender 0.08; Bid rigging 0.40; Dominant leader 0.04; Abuse of dominant position 0.03; Guaranteed buy‑backs 0.07; Exclusive dealing contracts 0.18.
  - Table counts: # Cartels 174; # Cartel members 1,037.
- Sectoral coverage:
  - Cartels detected across manufacturing, construction, wholesale and retail, transportation, and services.
  - No convicted firms in Agriculture and Education during 1994–2007 (together 0.5% of total value‑added).
  - Selected sector statistics (averages 1994–2007): e.g., NAF 15-16 Food products: Sales Share 0.0553, VA Share 0.0534, # Cartels 3, # Colluding Firms 19; NAF 45 Construction: Sales Share 0.0596, VA Share 0.0758, # Cartels 7, # Colluding Firms 42; NAF 50-52 Wholesale and retail: Sales Share 0.3518, VA Share 0.1872, # Cartels 11, # Colluding Firms 69; NAF 60-63 Transport and storage: Sales Share 0.0472, VA Share 0.0552, # Cartels 5, # Colluding Firms 27.
- Anticompetitive vs competitive firm summary statistics (1994–2007):
  - Market Share (%): Anticompetitive mean 3.43, Std. Dev. 10.79, Min 0, Max 100; Competitive mean 0.07, Std. Dev. 0.92, Min 0, Max 100.
  - Sales (thousands of euros): Anticompetitive mean 295,277; Competitive mean 207,056.
  - Value‑added (thousands of euros): Anticompetitive mean 118,799; Competitive mean 14,206.
  - ln Labor Productivity: Anticompetitive mean 3.87; Competitive mean 8.36.
  - Labor (employees): Anticompetitive mean 140; Competitive mean 121.
  - ln Wage: Anticompetitive mean 3.6; Competitive mean 7.45.
  - ln Capital/Labor ratio: Anticompetitive mean 2.25; Competitive mean 4.47.
  - Intermediates (thousands of euros): Anticompetitive mean 181,175; Competitive mean 147,945.
  - # Obs.: Anticompetitive 10,721; Competitive 12,441,919. # Firms: Anticompetitive 907; Competitive 2,167,168. # Exporters: Anticompetitive 613; Competitive 232,316.
- Key empirical comparisons and regression findings (Table A2):
  - After controlling for 2-digit sector and 4-digit industry fixed effects, anticompetitive firms:
    - have about 1900% more sales than competitive firms,
    - have a market share higher by 4 percentage points,
    - have 1150% more employment,
    - have 37% higher labor productivity.
  - Robustness: results hold when restricted to price fixing cartels (≈ 47 of 174 cartels) and when focusing on top producers.
- Cartel internal composition — productivity dispersion (2007):
  - Median ln labor productivity: within-industry mean 3.765 (Std. Dev. 0.450, IQ Range 0.482); within-cartel mean 4.474 (Std. Dev. 0.935, IQ Range 1.133).
  - IQ range: within-industry mean 0.722 (Std. Dev. 0.316); within-cartel mean 0.389 (Std. Dev. 0.347).
  - 90-10 percentile range: within-industry mean 1.463; within-cartel mean 0.531.
  - Interpretation: cartel members are relatively homogeneous compared to within-industry dispersion.

### Model environment, firm behavior, and collusion primitives
- Preferences and aggregation:
  - Representative household: E ∑_{t=0}^∞ β^t U(c_t, 1−l_t) with intra-temporal condition −U_{l,t}/U_{c,t} = W_t/P_t.
  - Final consumption c is CES across sectors with elasticity η: c = [∫_0^1 y_s^{(η−1)/η} ds]^{η/(η−1)}; sectoral inverse demand P_s/P = (y_s/c)^{−1/η}; price index P = [∫_0^1 P_s^{1−η} ds]^{1/(1−η)}.
  - Within sector s, y_s aggregates firm outputs q_{sk} with elasticity ρ: y_s = [∑_{k=1}^{K_s} q_{sk}^{(ρ−1)/ρ}]^{ρ/(ρ−1)}; within-sector inverse demands P_{sk}/P_s = (q_{sk}/y_s)^{−1/ρ}; sectoral price index P_s = [∑_{k=1}^{K_s} P_{sk}^{1−ρ}]^{1/(1−ρ)}.
  - Assumption: ρ < ∞ and 1 < η < ρ.
- Non-cartel (competitive) firms:
  - First-order condition pins markup μ_{si} = ε_{si}(ω_{si}) / (ε_{si}(ω_{si}) − 1) with ε_{si}(ω_{si}) = [1/ρ + (1/η − 1/ρ) ω_{si}]^{−1}, where ω_{si} is firm i’s sectoral revenue share.
  - Larger ω_{si} → higher market power → higher markups.
- Cartel members:
  - Distorted objective: π^C_k = π_k + ∑_{j∈C\{k\}} κ_{kj} π_j.
  - Effective demand elasticity under collusion: ε^C_{sk}(ω_{sk}) = [1/ρ + (1/η − 1/ρ) (ω_{sk} + ∑_{j∈C\{k\}} κ_{kj} ω_{sj})]^{−1}.
  - Collusion raises cartel members’ markups by internalizing effects on other cartel members.
- Collusion variants:
  - Benchmark (no collusion): κ_{kj} = 0 ∀ k, j (Nash‑Cournot heterogeneous firms).
  - Symmetric collusion: constant κ across pairs; increases weight on cartel market power.
  - Full collusion: κ_{kj} = 1 ∀ k, j ∈ C (joint profit maximization); cartel markup depends on combined market share ∑_{j∈C} ω_{sj}.
- Markup and price comparative statics:
  - First-order log change in cartel firm markup when ∆κ > 0:
    - ˆμ^C_{sk} = Υ_{sk} ˆP_s + (1/(ρ−1)) Υ_{sk} ω_{sk} (ω_{sC} − ω_{sk}) ∆κ.
    - Υ_{sk} := ω_{sk} (ρ−1) (1/η − 1/ρ) μ_{sk} / [1 + ω_{sk} (ρ−1) (1/η − 1/ρ) μ_{sk}] ∈ (0, 1).
  - Proposition 1 (symmetric collusion from competitive equilibrium):
    i) increases sectoral price index;
    ii) increases markups of all firms;
    iii) for cartel members, markup increase declines with firm size;
    iv) for non-cartel members, markup increase increases with firm size.
  - Corollary 1: market distortions increase with collusion intensity ∆κ and cartel market share ω_{sC}; sectoral price change given by equation (18).
  - Corollary 2: non-cartel members gain market shares via umbrella pricing; cartel members typically raise prices more than sectoral price increase, with overcharge larger for smaller cartel members under symmetric collusion.
- Aggregate productivity and markups:
  - Aggregation: Y = A L with A = [∫_0^1 (∑_{k=1}^{K_s} y_{sk}/Y z_{sk}^{−1}) ds]^{−1} (quantity-weighted harmonic average of firm productivities).
  - Aggregate markup μ_{agg} = P W / A is revenue-weighted harmonic mean of firm-level markups.
  - Sector productivity z_s and alternative efficient productivity A_{Efficient} defined in equations (21) and (22).
- Collusion channels affecting productivity:
  - Direct price effect: markup-dispersing price increases reduce sectoral productivity.
  - Demand reallocation effect: if cartels are top producers, demand shifts toward less productive firms → sectoral productivity decreases; French cartels empirically tend to be top producers.

### Cartel overcharge, parameterization, and estimation
- Empirical context on overcharges:
  - Connor and Bolotova (2006): median overcharge 19%, mean 29% (395 episodes).
  - Laborde (2019): median 15% (239 cases, France has 46 cases).
  - Laborde (2021): median 10%.
  - Levenstein et al. (2015): chemical cartels show large post-breakup price drops.
  - Asker et al. (2019): OPEC overcharge > 700% in the 1980s.
- Limitations for direct validation in this study:
  - No firm-level price data available.
  - Cartel breakups exogeneity unclear in decision files; causes often vague/unreported.
  - Focus: model feature of supracompetitive markups rather than pre/post price documentation.
- Key parameters determining aggregate productivity variation:
  - ρ (within-sector elasticity), η (across-sector elasticity), κ (collusion intensity).
  - Productivity z drawn Pareto with shape ξ; number of firms per sector geometric with parameter σ; cartel membership geometric with parameter ζ.
  - Time period = 1 year.
- Assigned/calibrated parameter values (Table 5):
  - β = 0.96; ψ = 1; δ = 0.1; α = 1/3; κ = 0.71 (estimated); ρ = 11.23; η = 1.50; ξ = 7.32; σ = 0.003; ζ = 0.15.
- Estimation of κ via inverse-markup regression:
  - Regression: Wl_sk / p_sk y_sk = a_0 + a_1 ω_sk + a_2 ∑_{j∈C} ω_sj + e_sk with a_0 = (ρ−1)/ρ, a_1 := (1−κ)(η−ρ)/(ηρ), a_2 := κ(η−ρ)/(ηρ).
  - Column 1 results: intercept = 0.70; own market share coeff = −0.53 (both significant at 1%); implied κ ≈ 0.70 (Table A7, # Observations 2,235).
  - Robustness: Brooks et al. (2021) obtain κ = 0.3 using different samples; authors test alternative κ values including 0.3.
- Calibration targets and fit:
  - Targets include aggregate markup μ_agg = 1.2; ratio (a_1 + a_2)/a_0 = −0.63; pooled distribution of relative sales; median # firms per 4-digit industry = 237; median # cartel members = 3.
  - Calibrated values: ρ = 11.2; η = 1.50; ξ = 7.32; σ = 0.003; ζ = 0.15.
  - Model reproduces targeted moments exactly where listed (e.g., Aggregate markup Data 1.2 → Model 1.2; Median # firms 237 → Model 237; Median # members per cartel 3 → Model 3).
  - Non-targeted moments (Table 7): Cartel premium (sales) Data 4.040 → Model 3.214; Cartel premium (employment) Data 3.306 → Model 3.006; Cartel premium (labor productivity) Data 0.478 → Model 0.208; Cartel premium (market share) Data 4.400 → Model 5.750; Median cartel overcharge (in %): Literature 10–20 → Model 12.5%.
- Markup distribution and κ role:
  - Ninety-ninth percentile markup: benchmark 1.34 vs competitive 1.15.
  - Ninetieth percentile sectoral markup: benchmark 1.31 vs competitive 1.26.
  - Increasing κ shifts cartel-member markup distribution rightward and more skewed; distributions shown for κ = 0, κ = 0.3, κ = 0.71.

### Aggregate productivity and welfare gains from dismantling cartels (key quantitative results)
- Aggregate productivity (Section 6.1):
  - Removing all cartels → aggregate productivity increase = 2.01% (∆TFP competitive Nash-Cournot).
  - Eliminating markup dispersion entirely (cartel equilibrium → efficient allocation) → increase = 5.41%.
  - Removing all cartels brings economy 37.07% closer to efficient allocation (Distance to efficient allocation = −37.07%).
  - Breaking down large cartels (cumulated market share > median): ∆TFP = 1.52%; distance to efficient allocation = −28.09%.
  - Breaking down smaller cartels: ∆TFP = 0.48%; distance = −8.91%.
  - Larger cartels charge higher markups and their dismantling reassigns resources toward more productive firms.
- Aggregate welfare (Section 6.2):
  - Eliminating all cartels decreases aggregate markup by −1.88 points and yields consumption-equivalent welfare gain ∆Welfare = 3.50%.
  - Breaking down large cartels: ∆Aggregate markup = −1.4 points; ∆Welfare = 2.60%.
  - Eliminating smaller cartels: ∆Aggregate markup = −0.51 points; ∆Welfare = 0.84%.
- Macro responses from eliminating cartels (Table 8, All cartels):
  - ∆C (in %) = 4.39; ∆K (in %) = 6.35; ∆Y (in %) = 4.77; ∆L (in %) = 0.98.
- Comparison with Harberger (1954) (Section 6.3):
  - Benchmark ∆TFP 2.01% vs sectoral aggregation ∆TFP 0.17% and unit elastic demand ∆TFP 0.2%; heterogeneity and non-unitary elasticities amplify effects relative to Harberger.
- Umbrella pricing costs (Section 6.4, Table 10):
  - Removing umbrella pricing has small quantitative effect: e.g., ∆TFP competitive Nash-Cournot Benchmark 2.01% vs No umbrella pricing 2.05% (difference 0.04 pp); ∆Welfare Benchmark 3.5% vs No umbrella pricing 3.52% (difference 0.02 pp).
  - Endogenous response of non-cartel members attenuates but does not eliminate gains.
- Gains from reducing collusion intensity κ (Section 6.5, Table 11):
  - Benchmark κ = 0.71: ∆TFP competitive Nash-Cournot = 2.01%; ∆Welfare = 3.5%.
  - κ→0.1: ∆TFP = 1.75%; ∆Welfare = 2.87%.
  - κ→0.2: ∆TFP = 1.44%; ∆Welfare = 2.26%.
  - κ→0.3: ∆TFP = 1.12%; ∆Welfare = 1.70%.
  - Reducing κ from 0.71 to 0.3 brings economy roughly 21% closer to efficient allocation.
- Cartel formation and stability (Section 6.6):
  - For low κ, almost all cartels profitable; for high κ, aggregate cartel profits can be lower than baseline.
  - Participation and incentive-compatibility constraints tighter for larger firms; small firms may require side-payments or non-monetary incentives to join.

### Robustness and sensitivity analyses
- Heterogeneous κ (Section 7.1, Table 12):
  - Allowing κ ~ truncated Normal across cartels (median κ across experiments 0.63–0.89) yields ∆TFP elimination ≈ 1.9%–3%; ∆Welfare ≈ 3.1%–4.7%.
  - Representative entries (σ^2_N = 1): Overcharge target 10% → ∆TFP 2.11%, ∆Welfare 3.36%, Median κ 0.81; Overcharge 15% → ∆TFP 2.95%, ∆Welfare 4.68%, Median κ 0.83; Overcharge 20% → ∆TFP 2.70%, ∆Welfare 4.45%, Median κ 0.87.
- Alternative robustness tests (Section 7.2):
  - Alternative aggregate markup targets (μ_agg = 1.1 and 1.3): ∆TFP competitive Nash‑Cournot = 1.61% and 2.13%; ∆Welfare = 2.41% and 4.33%.
  - Alternative median cartel size (4 members): ∆TFP = 2.47%; ∆Welfare = 4.33%.
  - Bertrand competition variant (price competition): ∆TFP competitive Nash‑Cournot = 0.39%; ∆Welfare = 1.16% (cartelized economy closer to efficient allocation under Bertrand).
  - Sensitivity to ρ (within-sector elasticity): low ρ = 5 → ∆TFP = 1.72%, ∆Welfare = 4.18%; high ρ = 20 → ∆TFP = 1.83%, ∆Welfare = 2.83%.
  - Sensitivity to κ: κ = 0.3 → ∆TFP = 0.81%, ∆Welfare = 1.65%; κ = 1 → ∆TFP = 2.66%, ∆Welfare = 4.46%.
  - Summary: benchmark elimination of cartels raises aggregate TFP by ≈ 2% and welfare by ≈ 3.5%; alternative calibrations produce ∆TFP competitive Nash‑Cournot spanning 0.39%–2.66% and ∆Welfare spanning 1.16%–4.46%.
- Policy implication emphasized: stronger antitrust enforcement and competition laws to break down cartels or reduce collusion intensity κ can yield sizeable aggregate productivity and welfare gains; complementary industrial and trade policies should be accompanied by robust competition policy.

### Data appendix, identification, and mathematical foundations
- Institutional history of French antitrust regulator summarized (1953 technical commission, 1963 law, 1977 Commission de la Concurrence, 1986 ordinance, 2001 leniency, 2008 Autorité de la Concurrence).
- Firm-level cartel dataset construction steps:
  - Scrape ADLC PDF decisions; Python text extraction; manual cleaning; match to SIREN via Orbis and Bing; verify with FICUS.
  - Final dataset fields: firm name, sales, decision case number, fine amount per firm, notification date, SIREN.
- Exact variable definitions preserved (APE Code, Capital, Colluder dummy, Employment, Gross Value‑Added, Labor Compensation, Market Shares at 4‑digit APE, Materials, NAF Code, Total Sales in thousands of euros, Wages, Market definition = 4‑digit industry).
- Identification and estimation highlights:
  - Parameter link: a1 + a2 = 1/ρ − 1/η; authors exploit projection of inverse markup on own and cartel market shares to identify κ and η given ρ.
  - Jacobian sensitivities: ρ, η, ξ, σ, ζ affect moments in interpretable ways (e.g., ρ ↑ increases relative demand for productive firms and aggregate markup).
- Mathematical appendix preserves exact model equations, first-order conditions, inverse markup expressions, special collusion cases (equations (36), (41), (42)), proof elements for Proposition 1, expressions for sectoral productivity change (equation (64)), overcharge expressions (equation (68)), participation and incentive constraints (equations (79), (83)), Bertrand variant (equations (86)–(88)), and consumption-equivalent welfare change formula ∆C = exp((WCartels − WCompetitive) (1 − β)) − 1.

*Source: Macroeconomic Effects of Market Structure Distortions: Evidence from French Cartels — Working Paper No. WP/2022/104 (content unit: wpiea2022104-print-pdf)*

### 2.1    Antitrust Decisions

### 2.1    Antitrust Decisions

### Role of the French Competition Authority (ADLC) and enforcement tools
- The French Competition Authority investigates and fines companies operating on the French market for anticompetitive practice: abuse of dominant position, collusion, or predatory pricing.
- Focus of this unit: collusion between firms (antocompetitive practices refers to collusion hereafter).
- Collusive behaviors include: trading information on prices, imposing standard form contracts, enforcing barriers to entry, imposing exclusive or selective distribution agreements, market sharing, purposely stepping down from calls for bids, or combinations thereof.
- Enforcement instruments:
  - Fines: set “according to the seriousness of the facts, the extent of the harm done to the economy, the individual situation of the company that has committed the infringement and of the group to which it belongs to, and whether it is an infringement that has been repeated or not”.
  - Fine caps: “10% of the global turnover of the group to which the company that is being fined belongs to” or at a maximum amount capped at 3 million euros if the infringement is committed by an entity other than a for-profit firm.
  - Injunctions: formal notifications to stop anticompetitive behavior.
- Cases spanning multiple countries are handled by the Directorate-General for Competition of the European Commission.

### Database construction and scope
- Source: investigation and decision PDF files published in French on the ADLC website; PDF files contain names of firms fined for anticompetitive practices.
- Extracted information: name of firms fined, amount of the fine, type of anticompetitive practice, duration of the practice, cause of breakup, year verdict returned, starting year of the investigation.
- Firms’ names and sales used to recover unique national identification code (“SIREN” code) from INSEE to match to firm-level production datasets.
- International cartels (cross-country) are excluded because market shares of foreign firms on the French market cannot be recovered; cross-country cartels are typically investigated by the European Commission (DG Comp).
- Dataset exclusions: cases where single firms were fined for abusing dominant position or repeat offenders are eliminated.
- Final dataset: 174 cartels and more than a thousand firms.
- Note: Cases notified by injunction alone were not used since these firms are often fined later and thus appear in the database.

### Administrative data matched to anticompetitive firms
- Matched using firms’ identification number.
- Data universe: French firms over the period 1994-2007; contains balance sheets and income statements of all French firms.
- Tax regimes included:
  - Regime of Normal Real Profits (BRN): firms with annual sales above 763K euros (230K euros for services).
  - Simplified Regime for the Self-Employed (RSI): firms selling at least 76.3K euros (but less than 763K euros) a year and more than 27K euros for services.
- BRN is the most relevant source: in 2003, BRN firms’ sales share in total sales was 94.3% and is constant over time.
- Administrative data allow recovery of firms’ market shares and variables used in empirical analysis.

### Characteristics of cartels: duration and size
- Average duration of a cartel: about 4.5 years.
- Median duration: about 3 years.
- Average number of firms per cartel: 6; median: 4.
- Standard deviation for number of firms per cartel: 7.
- Extremes: some cartels include more than 70 firms.
- Types of practices (prevalence noted in Table 1):
  - Price fixing: mean 0.35
  - Market allocation: mean 0.29
  - Production quotas: mean 0.04
  - Information sharing: mean 0.59
  - Repeat offender: mean 0.08
  - Bid rigging: mean 0.40
  - Dominant leader: mean 0.04
  - Abuse of dominant position: mean 0.03
  - Guaranteed buy-backs: mean 0.07
  - Exclusive dealing contracts: mean 0.18
- Table 1 counts:
  - # Cartels: 174
  - # Cartel members: 1,037
- Interpretation: communication/information sharing is pervasive; sharing customers and market shares helps sustain cartels.

### Cartels across sectors
- Detected cartels operate across manufacturing, construction, wholesale and retail, transportation, and services.
- Two sectors with no convicted firms over 1994-2007: Agriculture and Education (together account for 0.5% of total value-added).
- Selected sector statistics (columns represent NAF sector, sales share, VA share, # cartels, # colluding firms; averages over 1994-2007):
  - 15-16 Food products, beverages and tobacco: Sales Share 0.0553, VA Share 0.0534, # Cartels 3, # Colluding Firms 19
  - 45 Construction: Sales Share 0.0596, VA Share 0.0758, # Cartels 7, # Colluding Firms 42
  - 50-52 Wholesale and retail: Sales Share 0.3518, VA Share 0.1872, # Cartels 11, # Colluding Firms 69
  - 60-63 Transport and storage: Sales Share 0.0472, VA Share 0.0552, # Cartels 5, # Colluding Firms 27
  - (Full table lists sector-level shares and counts across NAF categories for 1994-2007.)

### Anticompetitive firms versus competitive firms (cartel premia)
- Summary statistics (period 1994-2007):
  - Market Share (%): Anticompetitive Firms mean 3.43, Std. Dev. 10.79, Min 0, Max 100; Competitive Firms mean 0.07, Std. Dev. 0.92, Min 0, Max 100.
  - Sales: Anticompetitive Firms mean 295,277; Competitive Firms mean 207,056 (sales in thousands of euros).
  - Value-added: Anticompetitive Firms mean 118,799; Competitive Firms mean 14,206 (in thousands of euros).
  - ln Labor Productivity: Anticompetitive Firms mean 3.87; Competitive Firms mean 8.36.
  - Labor (employees): Anticompetitive Firms mean 140; Competitive Firms mean 121.
  - ln Wage: Anticompetitive Firms mean 3.6; Competitive Firms mean 7.45.
  - ln Capital/Labor ratio: Anticompetitive Firms mean 2.25; Competitive Firms mean 4.47.
  - Intermediates: Anticompetitive Firms mean 181,175; Competitive Firms mean 147,945 (in thousands of euros).
  - # Obs.: Anticompetitive 10,721; Competitive 12,441,919
  - # Firms: Anticompetitive 907; Competitive 2,167,168
  - # Exporters: Anticompetitive 613; Competitive 232,316
- Key comparisons and findings:
  - Colluding firms have a much higher market share: 3.4% versus 0.07%.
  - Colluding firms sell more, spend more on intermediate goods, have more employees, are more capital-intensive, are more productive (labor productivity), and are more likely to be exporters.
  - Regression evidence (Table A2): after controlling for 2-digit sector and 4-digit industry fixed effects, anticompetitive firms:
    - have about 1900% more sales than competitive firms,
    - have a market share higher by 4 percentage points,
    - have 1150% more employment,
    - have 37% higher labor productivity.
  - Robustness: results hold when restricting to price fixing cartels (about 47 out of 174 cartels) and when focusing on top producers.

### Cartel internal composition: productivity dispersion
- Within-industry (non-cartel) versus within-cartel productivity moments (2007):
  - Median ln labor productivity:
    - Within-industry (no cartels) mean 3.765, Std. Dev. 0.450, IQ Range 0.482
    - Within cartel mean 4.474, Std. Dev. 0.935, IQ Range 1.133
  - IQ range (interquartile range):
    - Within-industry mean 0.722, Std. Dev. 0.316, IQ Range 0.250
    - Within cartel mean 0.389, Std. Dev. 0.347, IQ Range 0.666
  - 90-10 percentile range:
    - Within-industry mean 1.463, Std. Dev. 0.550, IQ Range 0.503
    - Within cartel mean 0.531, Std. Dev. 0.527, IQ Range 0.861
  - 95-5 percentile range:
    - Within-industry mean 1.971, Std. Dev. 0.699, IQ Range 0.675
    - Within cartel mean 0.572, Std. Dev. 0.540, IQ Range 0.945
- Interpretation:
  - Productivity differences across cartel members are smaller than those across competitive firms within narrowly defined industries.
  - Example: within-industry interquartile range about 0.72 (75th percentile about twice the productivity of 25th percentile), whereas within-cartel interquartile ratio is about 1.4-to-1.
  - Implication: firms within a cartel are relatively homogeneous, consistent with theoretical arguments that large cost differences across cartel members make collusion harder to sustain.

### Model framework overview
- A static, closed-economy granular model with heterogeneous firms; firms choose markups endogenously (following Atkeson and Burstein, 2008).
- Cartels coexist with competitive firms; model allows for Cournot and Bertrand competition.
- Economy: continuum of sectors, but in each sector only a finite number of firms compete.
- In equilibrium, firms’ endogenous markups increase with their market share.
- Collusion defined per Harrington Jr (2017): firms coordinate behavior to produce a supracompetitive outcome.
- Collusion alters firms’ internalization of sectoral output and price effects; collusion resembles cross-ownership distortions.
- Model nests several modes of collusion via a single parameter and provides micro-foundations to quantify aggregate productivity gains from eliminating cartels.
- Main model implication: collusion unambiguously raises prices and is harmful to consumers.

*Source: 2.1 Antitrust Decisions — wpiea2022104-print-pdf*

### 4.1    Environment

### 4.1    Environment

### Market Structure
- Representative household maximizes time-separable utility E ∑_{t=0}^∞ β^t U(c_t, 1−l_t) with intra-temporal condition −U_{l,t}/U_{c,t} = W_t/P_t.
- Final consumption c is produced from continuum of sector outputs y_s with CES aggregation (elasticity η):
  - c = [∫_0^1 y_s^{(η−1)/η} ds]^{η/(η−1)} (equation (2)).
  - Sectoral inverse demand: P_s/P = (y_s/c)^{−1/η} (equation (3)).
  - Price index: P = [∫_0^1 P_s^{1−η} ds]^{1/(1−η)} (equation (4)).
- Each sector s has K_s firms; sectoral output y_s is CES aggregation of firm outputs q_{sk} with elasticity ρ:
  - y_s = [∑_{k=1}^{K_s} q_{sk}^{(ρ−1)/ρ}]^{ρ/(ρ−1)} (equation (5)).
  - Within-sector inverse demands: P_{sk}/P_s = (q_{sk}/y_s)^{−1/ρ} (equation (6)).
  - Sectoral price index: P_s = [∑_{k=1}^{K_s} P_{sk}^{1−ρ}]^{1/(1−ρ)} (equation (7)).
- Assumption on elasticities: ρ < ∞ and 1 < η < ρ (goods more substitutable within than between sectors).

### Non-Cartel Members
- Competitive (non-cartel) firm i (i ∉ C_s) with linear labor costs and productivity z_{si} solves:
  - max_{q_{si}} (P_{si} − W/z_{si}) q_{si} (equation (8)) subject to inverse demand (equation (9)).
- Equilibrium price is a markup μ_{si} over marginal cost pinned down by idiosyncratic demand elasticity ε_{si}:
  - μ_{si} = ε_{si}(ω_{si}) / (ε_{si}(ω_{si}) − 1)
  - ε_{si}(ω_{si}) = [1/ρ + (1/η − 1/ρ) ω_{si}]^{−1} (equation (10)).
- ω_{si} := P_{si} q_{si} / ∑_{j=1}^{K} P_{sj} q_{sj} is firm i’s sectoral revenue share.
- Larger equilibrium market shares → more market power → higher markups. Demand elasticity is a harmonic weighted average of within- and between-sector elasticities.

### Cartel Members
- Cartel member k maximizes a distorted objective that internalizes other cartel members:
  - π^C_k = π_k + ∑_{j∈C\{k\}} κ_{kj} π_j (equation (11)).
- Cartel maximization:
  - max_{q_{sk}} [ (P_{sk} − W/z_{sk}) q_{sk} + ∑_{j∈C\{k\}} κ_{kj} (P_{sj} − W/z_{sj}) q_{sj} ] for all k ∈ C_s (equation (12)) subject to inverse demand (equation (13)).
- Collusion lowers effective demand elasticities for cartel members:
  - ε^C_{sk}(ω_{sk}) = [1/ρ + (1/η − 1/ρ) (ω_{sk} + ∑_{j∈C\{k\}} κ_{kj} ω_{sj})]^{−1} (equation (14)).
- Collusion raises cartel members’ markups because they internalize effects on other cartel members’ profits.

### Types of Collusion (model variants)
- Benchmark competitive economy:
  - κ_{kj} = 0 ∀ k, j → Nash-Cournot heterogeneous firms; markups follow equation (10). Used as counterfactual for aggregate gains from eliminating cartels.
- Cartels with symmetric collusion:
  - Symmetric κ across pairings → markups:
    - μ^C_{sk} = [ρ^{−1}/ρ + (η−1)/η − ρ^{−1}/ρ (ω_{sk} + κ ∑_{j∈C\{k\}} ω_{sj})]^{−1} (equation (15)).
  - Collusion intensity κ increases weight on cartel market power.
- Cartels with full collusion:
  - κ_{kj} = 1 ∀ k, j ∈ C → firms maximize joint profits; cartel member markup:
    - μ^C_{sk} = [ρ^{−1}/ρ + (η−1)/η − ρ^{−1}/ρ ∑_{j∈C} ω_{sj}]^{−1} (equation (16)).
  - All colluding firms in C charge the same markup governed by combined market share ∑_{j∈C} ω_{sj}; reduces within-cartel markup dispersion.

### How Collusion Distorts the Market Structure — Markups and Prices
- First-order log change in markups for cartel firm k when collusion intensity increases by ∆κ:
  - ˆμ^C_{sk} = Υ_{sk} ˆP_s (Umbrella Pricing) + (1/(ρ−1)) Υ_{sk} ω_{sk} (ω_{sC} − ω_{sk}) ∆κ (Cartel Overcharge) (equation (17)).
  - Υ_{sk} := ω_{sk} (ρ−1) (1/η − 1/ρ) μ_{sk} / [1 + ω_{sk} (ρ−1) (1/η − 1/ρ) μ_{sk}] ∈ (0, 1).
  - ˆP_s is percentage change of the sectoral price index.
- Interpretation:
  - Umbrella pricing term (Υ_{sk} ˆP_s) affects all firms in the sector.
  - Cartel overcharge term is specific to cartel members and increases with cartel market share ω_{sC} and collusion intensity ∆κ.
- Proposition 1 (Prices and Markups under Collusion):
  - Starting from the competitive equilibrium, symmetric collusion:
    i) increases the sectoral price index,
    ii) increases the markups of all firms,
    iii) for cartel members, the markup increase declines with firm size,
    iv) for non-cartel members, the markup increase increases with firm size.
- Corollary 1 (Intensive and Extensive Margins of Collusion):
  - Market distortions increase with i) collusion intensity ∆κ and ii) cartel market share ω_{sC}.
  - Sectoral price increase:
    - ˆP_s = (1/(ρ−1)) (1 / (1 − ∑_k ω_{sk} Υ_{sk})) ∑_{k∈C} Υ_{sk} (ω_{sC} − ω_{sk}) ∆κ (equation (18)).
- Corollary 2 (Market Shares under Collusion):
  - Non-cartel members gain market shares.
  - Evolution of market shares:
    - ˆω_{sk} = (ρ−1) (1 − Υ_{sk}) ˆP_s
    - ˆω^C_{sk} = (ρ−1) (1 − Υ_{sk}) ˆP_s − Υ_{sk} ω_{sk} (ω_{sC} − ω_{sk}) ∆κ (equation (19)).
  - Non-colluding firms raise prices via umbrella pricing but less than full sectoral price increase, allowing them to gain market shares. Colluding firms tend to raise prices more than sectoral price increase; overcharge larger for smaller cartel members under symmetric collusion.

### Illustration: Markups, Prices, and Harberger Triangles
- In competitive equilibrium:
  - More productive firms have lower prices but charge higher markups due to larger market shares.
  - Sectoral price index is lower than individual prices due to love of variety.
- After cartel formation:
  - Cartel members’ prices and markups increase; sectoral price index rises.
  - Non-cartel members raise markups and prices via umbrella pricing; gains differ by productivity and cartel membership.
- Harberger triangles with firm heterogeneity:
  - Deadweight losses form a collection of Harberger triangles across heterogeneous firms.
  - More productive firms generate larger welfare losses when exercising market power.
  - Markup dispersion ranges from μ_η := η/(η−1) for dominant firms (market share → 1) to μ_ρ := ρ/(ρ−1) for small firms (market share → 0).
  - Collusion increases sizes of Harberger triangles for cartel members; effect on non-cartel deadweight loss ambiguous (markup up but output up).

### Aggregate Productivity
- Aggregation yields output Y = A L with aggregate productivity A and total labor L.
- Aggregate productivity expression:
  - A = [∫_0^1 (∑_{k=1}^{K_s} y_{sk}/Y z_{sk}^{−1}) ds]^{−1} (equation (20)).
  - A is a quantity-weighted harmonic average of firm productivities.
- Aggregate markup μ_{agg} = P W / A is a revenue-weighted harmonic mean of firm-level markups:
  - μ_{agg} = [∫_0^1 (∑_{k=1}^{K_s} p_{sk} y_{sk} / (P Y) μ_{sk}^{−1}) ds]^{−1}.
- Alternative representation:
  - A = [∫_0^1 (μ_{agg}/μ_s)^η z_s^{η−1} ds]^{1/(η−1)} where sector productivity z_s:
    - z_s = [∑_{k=1}^{K_s} (μ_s/μ_{sk})^ρ z_{sk}^{ρ−1}]^{1/(ρ−1)} (equation (21)).
- Counterfactuals:
  - Compare A_{Cartel} with A_{Competitive} (Nash-Cournot).
  - Efficient productivity (no markup dispersion):
    - A_{Efficient} = (∫_0^1 (∑_{k=1}^{K_s} z_{sk}^{ρ−1})^{(η−1)/(ρ−1)} ds)^{1/(η−1)} (equation (22)).

### Collusion and Productivity
- Change in sectoral productivity ˆz_s:
  - ˆz_s = ∑_k ω_{sk} (μ_s/μ_{sk} − 1) ˆP_{sk} + (ρ−1) ∑_k ω_{sk} μ_s/μ_{sk} (ˆP_{sk} − ˆP_s) (equation (23)).
- Two channels for collusion impact:
  - Direct price effect: with markup dispersion, price increases by high-markup firms reduce sectoral productivity.
  - Demand reallocation effect: if cartels consist of top producers with above-average markups, demand shifts toward less productive firms → sectoral productivity decreases. Conversely, if cartels consist of smallest firms, demand shifts to larger firms → productivity may increase.
- Empirical note from text: French cartels tend to be made of top firms, implying a negative relationship between aggregate productivity and collusion intensity (illustrated in Figure A6 referenced in the source).

### Collusion and Welfare
- Extend model with capital accumulation and elastic labor supply to evaluate welfare (following Edmond et al., 2015, 2018).
- Aggregate markup acts as a distortionary wedge: an increase in μ_{agg} due to cartelization reduces aggregate scale of production and decreases representative consumer welfare.
- Welfare change computed in consumption-equivalent units accounting for transitional dynamics to new steady state (details in Appendix E.7).

*Source: wpiea2022104-print-pdf - 4.1    Environment*

### 4.4    Cartel Overcharge

### 4.4    Cartel Overcharge

### Empirical relevance of cartel overcharges
- Meta-analysis evidence (Connor and Bolotova (2006)) on 395 cartel episodes over 250 years: median overcharge is 19% and mean overcharge is 29%.
- Laborde (2019): median cartel overcharge is 15% in a sample of 239 cases judged by European national competition authorities and the European Commission; France has 46 cases.
- Laborde (2021): median cartel overcharge is 10%.
- Levenstein et al. (2015): analysis of seven chemical cartels shows each cartel breakdown was followed by a large price drop.
- Asker et al. (2019): OPEC cartel overcharge higher than 700% in the 1980s.
- The model generates cartel overcharges via changes in the collusion intensity parameter κ.

### Limitations for direct empirical validation in this study
- No firm-level price data are available in the authors’ dataset.
- Market concentration is the outcome of many supply and demand forces; cartel breakdowns may be endogenous to industry performance.
- Exogenous cartel breakdowns cannot be cleanly inferred from decision files: causes of breakdowns are often vague or unreported.
- Consequence: the paper focuses on a model feature—supracompetitive markups and distorted price schedules—rather than directly documenting pre/post cartel concentration and price changes.

### Model role and key parameters
- The key parameters determining aggregate productivity variation in presence of cartels are:
  - ρ: within-sector elasticity of substitution
  - η: across-sector elasticity of substitution
  - κ: collusion intensity parameter
- Interpretation:
  - The gap between ρ and η determines how dispersion in market shares translates into markup dispersion.
  - κ governs the extent to which cartel members internalize effects on other cartel members; higher κ leads to larger estimates of cartel impact.

### Parameterization details
- Productivity distribution: firms within a sector draw productivity z from a Pareto distribution with shape parameter ξ.
- Number of firms per sector: drawn from a geometric distribution with parameter σ; probability of K_s firms is σ(1−σ)^{K_s−1}; average number of firms is 1/σ.
- Number of cartel members: at most one cartel per sector; when cartelized, members are top producers; number drawn from geometric distribution with parameter ζ; average cartel size is 1/ζ.
- Time period: one year.
- Assigned parameter values (Table 5):
  - β (Discount factor) = 0.96
  - ψ (Labor supply elasticity) = 1
  - δ (Capital depreciation rate) = 0.1
  - α (Output elasticity of capital) = 1/3
  - κ (Collusion intensity) = 0.71 (estimated)
  - ρ (Substitution within sectors) = 11.23
  - η (Substitution between sectors) = 1.50
  - ξ (Pareto shape parameter) = 7.32
  - σ (Geometric parameter firms) = 0.003
  - ζ (Geometric parameter cartel members) = 0.15
- Notes: parameters chosen to minimize distance between model and data moments from French micro data in 2007.

### Estimation of collusion intensity κ
- Model equilibrium inverse markup equation implies regression:
  Wl_sk / p_sk y_sk = a_0 + a_1 ω_sk + a_2 ∑_{j∈C} ω_sj + e_sk
  where a_0 = (ρ−1)/ρ, a_1 := (1−κ)(η−ρ)/(ηρ), a_2 := κ(η−ρ)/(ηρ).
- Using firm labor share as proxy for markup (no price data):
  - Column 1 regression (colluding firms without joint market share): intercept = 0.70, own market share coeff = −0.53; both significant at 1%.
  - Column 2 (includes cartel joint market share): own market share coeff remains negative but becomes not significant; joint market share coeff is negative and significant at 1%; intercept remains positive and significant.
  - Column 3 (adds year fixed effects): point estimates and standard errors change little; estimated κ = 0.71.
- κ recovered by κ̂ = â_2 / (â_1 + â_2).
- Comparison and robustness:
  - Brooks et al. (2021) use a similar equation but find κ = 0.3; higher κ in this study attributed to using firms detected as colluding.
  - Robustness tests include alternative κ values (including 0.3) and allowing κ to vary across cartels to match cartel overcharge targets from IO literature.

### Calibration targets and fit strategy
- Parameters to estimate: θ ≡ {ρ, η, ξ, σ, ζ}.
- Objective: minimize model-data distance function summing absolute relative differences across moments.
- Moments used:
  - Aggregate markup target μ_agg = 1.2 (consistent with literature for France in 2007).
  - Regression-based restriction: ratio (a_1 + a_2)/a_0 targeted at −0.63 (from Table A7).
  - Distribution of relative sales (pooled across industries): target moments from French data (Panel B & C of Table 6).
  - Median number of firms per 4-digit industry targeted at 237.
  - Median number of cartel members targeted at 3.

### Calibrated parameter values and model fit (summary)
- Calibrated parameter values (from Table 5 bottom rows):
  - ρ = 11.2, η = 1.50, ξ = 7.32, σ = 0.003, ζ = 0.15.
- Model reproduces targeted moments (Table 6):
  - Aggregate markup: Data 1.2 → Model 1.2
  - Median # firms per sector: Data 237 → Model 237
  - Median # members per cartel: Data 3 → Model 3
  - Ratio ((a_1 + a_2)/a_0): Data −0.63 → Model −0.63
  - Distribution of relative sales: model matches the skew (examples)
    - Fraction firms with relative sales ≤ 0.10: Data 0.306 → Model 0.213
    - Fraction firms with relative sales ≤ 0.50: Data 0.646 → Model 0.688
  - Fraction of sales by firm size thresholds (Panel C examples):
    - ≤ 0.10: Data 0.012 → Model 0.019
    - ≤ 10: Data 0.185 → Model 0.190
- Non-targeted moments (Table 7): model reproduces several premiums and dispersions reasonably well
  - Cartel premium (sales): Data 4.040 → Model 3.214
  - Cartel premium (employment): Data 3.306 → Model 3.006
  - Cartel premium (labor productivity): Data 0.478 → Model 0.208
  - Cartel premium (market share): Data 4.400 → Model 5.750
  - Standard deviation log sales: Data 1.391 → Model 1.366
  - Median cartel overcharge (in %): Literature 10–20 → Model 12.5%

### Markup distribution and role of κ
- Benchmark markups: higher for top firms with cartels than in competitive economy.
  - Ninety-ninth percentile markup: benchmark 1.34 vs competitive 1.15.
  - Ninetieth percentile sectoral markup: benchmark 1.31 vs competitive 1.26.
- Effect of κ on cartel members’ markups (Figure 3):
  - κ = 0: markups not common across cartel members; more productive firms charge higher markups.
  - Increasing κ shifts cartel members’ markup distribution to the right and makes it more skewed, increasing misallocation.
  - The figure displays distributions for κ = 0, κ = 0.3, and κ = 0.71; median of each distribution shown.

### Aggregate gains from breaking down cartels (Table 8)
- Productivity gains (comparing cartel equilibrium to competitive Nash-Cournot and to efficient allocation):
  - ∆TFP competitive Nash-Cournot (in %): All cartels 2.01; Larger cartels 1.52; Smaller cartels 0.48
  - ∆TFP efficient allocation (in %): All cartels 5.41; Larger cartels 5.41; Smaller cartels 5.41
  - Distance to efficient allocation (in %): All cartels −37.07; Larger cartels −28.09; Smaller cartels −8.91
- Aggregate welfare and macro changes from eliminating cartels:
  - ∆Aggregate markup (in points): All cartels −1.88; Larger cartels −1.4; Smaller cartels −0.51
  - ∆C (in %): All cartels 4.39; Larger cartels 3.31; Smaller cartels 1.09
  - ∆K (in %): All cartels 6.35; Larger cartels 4.77; Smaller cartels 1.62
  - ∆Y (in %): All cartels 4.77; Larger cartels 3.60; Smaller cartels 1.19
  - ∆L (in %): All cartels 0.98; Larger cartels 0.73; Smaller cartels 0.27
  - ∆Welfare (in %): All cartels 3.50; Larger cartels 2.60; Smaller cartels 0.84
- Notes: efficient allocation corresponds to equilibrium without markup dispersion; columns 2 and 3 consider eliminating only cartels with cumulated market share higher (column 2) or lower (column 3) than the median cumulated market share of all cartels.

*Source: wpiea2022104-print-pdf - 4.4    Cartel Overcharge*

### 6.1    Aggregate Productivity Gains

### 6.1    Aggregate Productivity Gains

### Key findings
- Removing all cartels would increase aggregate productivity by 2%.
- Eliminating markup dispersion entirely (transition from the cartel equilibrium to the efficient allocation) would increase aggregate productivity by 5.4%.
- Removing all cartels would bring the economy 37% closer to the efficient allocation.
- Breaking down large cartels (cumulated market share > median) increases aggregate productivity by 1.5% (about three-quarters of total gain) and would bring the economy 28% closer to the efficient allocation.
- Breaking down smaller cartels increases aggregate productivity by 0.5% and would bring the economy 9% closer to the efficient allocation.
- Larger cartels charge higher markups; dismantling them reallocates relatively more resources toward more productive firms.

### Comparison note with Edmond et al. (2018)
- Edmond et al. (2018) find eliminating size-dependent markup dispersion increases aggregate productivity by 1%-3%.
- Going from the competitive oligopoly equilibrium to the efficient allocation in this model yields aggregate productivity gains of 3.4% (5.41% - 2.01%), consistent with Edmond et al.’s upper bound estimate of 3%.

---

### 6.2    Welfare Gains

### Aggregate markup and welfare effects
- Eliminating all cartels would decrease the aggregate markup by about 1.9 points.
- Given the model’s relationship between aggregate labor share and aggregate markup, changes in cartelization generate changes in the labor share.
- Breaking down all cartels would lead to a consumption-equivalent welfare gain of 3.5%.

### Breakdown by cartel size
- Breaking down large cartels would decrease the aggregate markup by 1.4 percentage points and translate into a 2.6% increase in welfare.
- Eliminating smaller cartels would decrease the aggregate markup by 0.5 percentage points and translate into a 0.8% increase in welfare.
- Targeting very large cartels yields sizeable aggregate welfare gains.

---

### 6.3    Comparison with Harberger (1954)

### Channels explaining larger estimates than Harberger
- Two channels amplify the impact relative to Harberger (1954): firm heterogeneity and non-unitary elasticities of substitution.
- Harberger uses sectoral data and assumes unit demand elasticities; this model uses firm-level heterogeneity and higher elasticities.
- Not accounting for within-sector heterogeneity and assuming unit demand elasticities reduces the estimated aggregate productivity gain from breaking down cartels to 0.17% (sectoral aggregation) and to 0.2% when also assuming unit elasticities.

### Illustration from Table 9
- ∆TFP competitive Nash-Cournot (in %) in benchmark: 2.01; sectoral aggregation: 0.17; unit elastic demand: 0.2.
- Results highlight importance of firm heterogeneity within sectors and appropriate demand elasticities.

---

### 6.4    Aggregate Costs of Umbrella Pricing

### Umbrella pricing effect
- When cartels are broken down, prices of all firms in cartelized industries fall; demand may reallocate toward less productive non-cartel members, dampening productivity gains.
- Table 10 comparison (Benchmark vs No umbrella pricing effect):
  - ∆TFP competitive Nash-Cournot (in %): Benchmark 2.01; No umbrella pricing effect 2.05; difference 0.04 (pp).
  - ∆TFP efficient allocation (in %): both 5.41.
  - Distance to efficient allocation (in %): Benchmark −37.07; No umbrella pricing effect −37.96; difference 0.89 (pp).
  - ∆Aggregate markup (in points): Benchmark −1.88; No umbrella pricing effect −1.71; difference −0.17 (pp).
  - ∆C (in %): Benchmark 4.39; No umbrella pricing effect 4.34; difference −0.05 (pp).
  - ∆K (in %): Benchmark 6.35; No umbrella pricing effect 6.14; difference −0.21 (pp).
  - ∆Y (in %): Benchmark 4.77; No umbrella pricing effect 4.70; difference −0.07 (pp).
  - ∆L (in %): Benchmark 0.98; No umbrella pricing effect 0.90; difference −0.08 (pp).
  - ∆Welfare (in %): Benchmark 3.5; No umbrella pricing effect 3.52; difference 0.02 (pp).
- The endogenous response of non-cartel members attenuates the impact of cartel breakdowns on aggregate productivity and welfare, but the quantitative effect is small.

---

### 6.5    Gains from Curbing the Collusion Intensity

### Policy-relevant margin: collusion intensity κ
- Instead of full dismantling, reducing collusion intensity κ (cartel members assign lower weight to each other’s profits) yields substantial gains.
- Table 11 (columns for κ→0.1, κ→0.2, κ→0.3; benchmark κ→0):
  - ∆TFP competitive Nash-Cournot (in %): Benchmark 2.01; κ→0.1: 1.75; κ→0.2: 1.44; κ→0.3: 1.12.
  - ∆TFP efficient allocation (in %): 5.41 in all columns.
  - Distance to efficient allocation (in %): Benchmark −37.07; κ→0.1: −32.31; κ→0.2: −26.63; κ→0.3: −20.71.
  - ∆Aggregate markup (in points): Benchmark −1.88; κ→0.1: −1.05; κ→0.2: −0.54; κ→0.3: −0.25.
  - ∆C (in %): Benchmark 4.39; κ→0.1: 3.40; κ→0.2: 2.56; κ→0.3: 1.86.
  - ∆K (in %): Benchmark 6.35; κ→0.1: 4.50; κ→0.2: 3.13; κ→0.3: 2.12.
  - ∆Y (in %): Benchmark 4.77; κ→0.1: 3.62; κ→0.2: 2.67; κ→0.3: 1.91.
  - ∆L (in %): Benchmark 0.98; κ→0.1: 0.55; κ→0.2: 0.29; κ→0.3: 0.13.
  - ∆Welfare (in %): Benchmark 3.5; κ→0.1: 2.87; κ→0.2: 2.26; κ→0.3: 1.70.

### Interpretation
- Aggregate productivity gains remain large (1.12% to 1.75%) and decrease in κ.
- Reducing κ from 0.71 to 0.3 brings the economy roughly 21% closer to the efficient allocation.
- A tougher antitrust environment that lowers κ yields substantial aggregate gains.

---

### 6.6    Cartel Formation and Stability

### Profits and heterogeneity
- For sufficiently low collusion intensity κ, almost all cartels are profitable (Figure 4a).
- For high κ, aggregate profits of cartel members can be lower than in the baseline.
- There is substantial heterogeneity across cartels; a right tail exists where cartels generate aggregate gains even at higher κ (Figure 4b).

### Participation and incentive-compatibility constraints
- Participation constraint (upper bound on overcharge Θ_sk when joining cartel), from Appendix E.5:
  - Θ_sk < [ρ−η / ρ−ε_sk − Υ_sk] ˆP^C_s
  - The bracketed term is decreasing with size; constraint is less binding for smaller firms (smaller firms gain from sectoral price increases).
- Incentive-compatibility (sustainability) constraint with discount factor δ:
  - Θ_sk < [ρ−η / ρ−ε_sk − Υ_sk][ˆP^C_s − (1−δ) ˆP^C\{k}_sk]
  - ˆP^C\{k}_sk is the price level increase if all other cartel members except firm k apply the overcharge.
  - This constraint is always tighter than the participation constraint and is not monotonically decreasing with size.
- Implications:
  - Larger firms face tighter incentive constraints due to larger price impact and higher temptation to free-ride.
  - Small firms may require side-payments, threats, or non-monetary incentives to participate.

---

### 7    Robustness

### 7.1    Heterogeneous κ

#### Experimental design
- Each cartel’s κ is drawn from a truncated Normal distribution on [0,1] with mean μ_N and variance σ^2_N.
- μ_N chosen so model-generated median cartel overcharge matches targets: 10%, 15%, 20%.
- Two variance values considered: σ^2_N = 1 and σ^2_N = 2.

#### Results summary (Table 12)
- Allowing heterogeneous κ yields aggregate productivity gains from eliminating cartels around 1.9%-3%, close to the 2% benchmark.
- Welfare gains range from 3.1% to 4.7%, around benchmark 3.5%.
- Median κ across experiments ranges from 0.63 to 0.89, close to baseline κ = 0.71.
- Representative entries (Panel A, σ^2_N = 1):
  - Overcharge target 10%: ∆TFP competitive Nash-Cournot (in %) 2.11; ∆TFP efficient allocation (in %) 7.01; Distance to efficient allocation (in %) −30.14; ∆Aggregate markup (in points) −0.91; ∆Welfare (in %) 3.36; Median κ 0.81.
  - Overcharge target 15%: ∆TFP competitive Nash-Cournot (in %) 2.95; ∆TFP efficient allocation (in %) 8.30; Distance −35.56; ∆Aggregate markup −1.06; ∆Welfare 4.68; Median κ 0.83.
  - Overcharge target 20%: ∆TFP competitive Nash-Cournot (in %) 2.70; ∆TFP efficient allocation (in %) 6.94; Distance −38.84; ∆Aggregate markup −1.56; ∆Welfare 4.45; Median κ 0.87.

#### Interpretation
- Assuming homogeneous κ appears conservative; allowing heterogeneity does not dramatically change aggregate estimates.
- Estimating κ from data or calibrating it to overcharge moments yields similar values for κ and similar aggregate gains.

*Source: wpiea2022104-print-pdf - 6.1    Aggregate Productivity Gains*

### 7.2    Alternative Robustness Tests

### 7.2    Alternative Robustness Tests

### Alternative aggregate markup targets
- Benchmark aggregate markup target set equal to 1.2; alternative targets considered: low = 1.1 and high = 1.3.
- Productivity gains from breaking down cartels under alternative markup targets (Panel A, columns 1–2):
  - ∆TFP competitive Nash-Cournot (in %): 1.61, 2.13
  - ∆TFP efficient allocation (in %): 4.87, 5.77
  - Distance to efficient allocation (in %): −33.06, −36.91
  - ∆Aggregate markup (in points): −0.29, −3.61
  - ∆C (in %): 2.64, 5.69
  - ∆K (in %): 2.96, 9.13
  - ∆Y (in %): 2.71, 6.32
  - ∆L (in %): 0.16, 1.72
  - ∆Welfare (in %): 2.41, 4.33
- Text summary: Targeting alternative aggregate markup values does not yield dramatic differences with benchmark estimates; productivity gains from breaking down cartels remain important, ranging from 1.6% to 2.1%. Consumption-equivalent welfare gains range from 2.4% to 4.3%.

### Alternative cartel target
- Benchmark median number of cartel members targeted = 3; alternative target = 4 (matching median number of firms operating in a cartel per Table 1).
- Results for alternative cartel target (Panel A, column 3):
  - ∆TFP competitive Nash-Cournot (in %): 2.47
  - ∆TFP efficient allocation (in %): 6.10
  - Distance to efficient allocation (in %): −40.48
  - ∆Aggregate markup (in points): −2.28
  - ∆C (in %): 5.38
  - ∆K (in %): 7.77
  - ∆Y (in %): 5.85
  - ∆L (in %): 1.19
  - ∆Welfare (in %): 4.33
- Text summary: With a median cartel size of four, productivity gains are 2.47% and consumption-equivalent welfare gain is 4.3%. More industries have larger cartel membership, increasing misallocation and the gains from eliminating cartels.

### Bertrand competition (price competition)
- Panel A, column 4 reports Bertrand competition results:
  - ∆TFP competitive Nash-Cournot (in %): 0.39
  - ∆TFP efficient allocation (in %): 2.42
  - Distance to efficient allocation (in %): −16.11
  - ∆Aggregate markup (in points): −2.00
  - ∆C (in %): 2.06
  - ∆K (in %): 4.15
  - ∆Y (in %): 2.47
  - ∆L (in %): 1.05
  - ∆Welfare (in %): 1.16
- Text summary:
  - Under Bertrand competition aggregate productivity gain is equal to 0.4% and is smaller than with Cournot (quantity) competition.
  - Model features considerably less markup dispersion under Bertrand; cartelized economy is closer to the efficient allocation (distance to efficient frontier reduced by about 16%).
  - The level of the aggregate markup decreases by 2 percentage points, translating into a consumption-equivalent welfare gain of about 1.2%.

### Sensitivity to ρ (elasticity of substitution within sectors)
- Benchmark ρ = 11.2; robustness tested at low ρ = 5 and high ρ = 20 (Panel B, columns 1–2).
- Panel B results:
  - Low ρ (column 1):
    - ∆TFP competitive Nash-Cournot (in %): 1.72
    - ∆TFP efficient allocation (in %): 4.94
    - Distance to efficient allocation (in %): −34.88
    - ∆Aggregate markup (in points): −4.76
    - ∆C (in %): 5.67
    - ∆K (in %): 9.78
    - ∆Y (in %): 6.36
    - ∆L (in %): 2.06
    - ∆Welfare (in %): 4.18
  - High ρ (column 2):
    - ∆TFP competitive Nash-Cournot (in %): 1.83
    - ∆TFP efficient allocation (in %): 5.28
    - Distance to efficient allocation (in %): −34.71
    - ∆Aggregate markup (in points): −0.63
    - ∆C (in %): 3.22
    - ∆K (in %): 3.92
    - ∆Y (in %): 3.37
    - ∆L (in %): 0.35
    - ∆Welfare (in %): 2.83
- Text summary: The model cannot match the aggregate markup target with extreme ρ values, but gains from dismantling cartels remain large, with aggregate productivity gains ranging from 1.72% to 1.83% and welfare gains from 2.83% to 4.18%.

### Sensitivity to κ (intensity of collusion)
- Benchmark κ = 0.71; alternative κ values tested: low κ = 0.3 and high κ = 1 (Panel B, columns 3–4).
- Panel B results:
  - Low κ (column 3, κ = 0.3):
    - ∆TFP competitive Nash-Cournot (in %): 0.81
    - ∆TFP efficient allocation (in %): 4.03
    - Distance to efficient allocation (in %): −20.10
    - ∆Aggregate markup (in points): −1.56
    - ∆C (in %): 2.37
    - ∆K (in %): 4.00
    - ∆Y (in %): 2.69
    - ∆L (in %): 0.81
    - ∆Welfare (in %): 1.65
  - High κ (column 4, κ = 1):
    - ∆TFP competitive Nash-Cournot (in %): 2.66
    - ∆TFP efficient allocation (in %): 6.22
    - Distance to efficient allocation (in %): −42.84
    - ∆Aggregate markup (in points): −1.79
    - ∆C (in %): 5.31
    - ∆K (in %): 7.18
    - ∆Y (in %): 5.68
    - ∆L (in %): 0.94
    - ∆Welfare (in %): 4.46
- Text summary:
  - When κ = 0.3, aggregate productivity and consumption-equivalent welfare increase by 0.8% and 1.65%, respectively, if firms behave competitively.
  - When κ = 1, aggregate gains are higher than benchmark; gains increase with κ because cartel members raise markups more when κ is higher, increasing misallocation.
  - Overall, gains from dismantling cartels remain sizeable across alternative κ values.

### Summary of robustness findings
- Core quantified implications are robust across a range of alternative calibrations:
  - Eliminating cartels raises aggregate TFP by about 2% and welfare by about 3.5% in the benchmark.
  - Alternative calibrations produce ∆TFP competitive Nash-Cournot values spanning 0.39 to 2.66 and ∆Welfare values spanning 1.16 to 4.46 across Table 13 experiments.
- Policy implication emphasized in the chapter: antitrust enforcement and competition laws aimed at breaking down cartels can yield sizeable gains; complementary policies (industrial policy, trade liberalization) should be accompanied by robust competition laws to prevent cartel formation and dampening of productivity gains.

*Source: Section 7.2, "Alternative Robustness Tests", from the provided PDF content.*

### Appendix D. In Appendix E we provide the mathematical proofs of the results pre-

### Appendix D. In Appendix E we provide the mathematical proofs of the results presented in the main text.

### A. Data Appendix — Institutional background and firm-level cartel database
- Institutional evolution of French antitrust regulator:
  - First established in 1953 as the French Technical Commission for Collusions and Dominant Positions (Décret n o 53-704 du 9 août 1953).
  - 1963: objectives extended to investigate dominant positions (Loi n o 63-628 du 2 juillet 1963).
  - 1977: became the Competition Commission (Commission de la Concurrence).
  - 1986 Ordinance: companies can directly refer cases; antitrust body gains independence and direct fining powers (except for merger projects).
  - 2001 New Economic Regulation Law: introduces leniency and transaction programs.
  - 2008: Competition Council → Competition Authority (Autorité de la Concurrence, ADLC); 2008 Law on the Modernization of Economy grants right to review mergers independently and to investigate on its own.
- Firm-level cartel dataset construction (period 1994-2019):
  - Step 1: Scrape ADLC decision PDF files (contain market situation, notification date, fined firms, types of infraction, sales, duration).
  - Step 2: Use Python textual analysis to extract firm names, sales, notification dates, fines; manual cleaning and double-checking against original PDFs.
  - Step 3: Match firms to administrative identifiers using Orbis Batch Search to recover SIREN numbers; complement with a Bing-search Python script; manually verify SIREN by comparing sales to FICUS.
  - Final dataset contains: firm name, sales, decision case number, fine amount per firm, notification date, and SIREN for matching to FICUS.

### A.3. Variables and market definition (exact definitions preserved)
- APE Code: 4-digit industry code. Before 2008, APE codes are available in a 4-digit format corresponding to the NAF Rev. 1 classification. Source: FICUS and authors’ calculation
- Capital: Net book value of capital. Deflated by sector-level price indices from EUKLEMS (Jäger, 2017). Source: FICUS and authors’ calculation
- Colluder: Dummy variable that takes the value one if the firm engaged in anti-competitive practices in a given year. Source: Moreau-Panon database
- Employment: Total number of employees. Source: FICUS
- Gross Value-Added: Directly available in FICUS; equal to total sales minus input expenses taking into account changes in inventories. Source: FICUS
- Labor Compensation: Sum of salaries and social benefits paid by the employer. Source: FICUS
- Market Shares: Defined at the 4-digit level; computed as firm’s domestic sales divided by total domestic sales in the same market. Source: FICUS and authors’ calculation
- Materials: Sum of expenditures on raw materials, final goods and other categories; deflated by 2-digit intermediate goods price indices from EUKLEMS. Source: FICUS and authors’ calculation
- NAF Code: 2-digit sector code according to NACE Rev. 1 classification. Source: FICUS
- Total Sales: Total sales (domestic sales plus export sales) reported by the firm in thousands of euros. Source: FICUS
- Wages: Labor compensation divided by employment. Source: FICUS and authors’ calculation
- Market definition: Relevant market = firm’s 4-digit industry code (Code APE). Sector definition follows NAF Rev. 1.

### B. Descriptive and regression table highlights (exact coefficients, samples, and moments)
- Table A1 (Cartels by Sector, 2007): sector sales and VA shares and counts reported per NAF; e.g., Wholesale and retail (50-52): Sales Share 0.3473, VA Share 0.1930, # Cartels 11, # Colluding Firms 122.
- Table A2 (Anticompetitive Firm Premium, period 1994-2007): coefficient on Collude for ln Sales:
  - Column examples: 4.040*** (SE 0.092); other specifications report 3.582*** (0.092), 3.002*** (0.082).
  - Market Share, ln Employment, ln Labor Productivity columns likewise report Collude coefficients (e.g., 4.400*** (0.548), 4.297*** (0.542), 4.028*** (0.473)).
  - # Obs. 12,452,544 across specifications; R2 reported (example: 0.002, 0.177, 0.315).
- Table A3 (Price-Fixing cartels subset, period 1994-2007): Collude ln Sales coefficients:
  - Examples: 3.912*** (0.149), 3.268*** (0.140), 2.881*** (0.124); # Obs. 12,450,922.
- Table A4 (Anticompetitive Firms and Firm Rank): ln Rank Market Share coefficients: -0.0003*** (0.0000); Top 4 Industry coefficients: 0.0163*** (0.0015); # Obs. 12,452,544.
- Table A5 (Log Sales Dispersion): Non-Cartel Members median 6.623 (Std. Dev. 1.264, IQ Range 1.561); Cartel Members median 0.845 (Std. Dev. 2.347, IQ Range 2.311).
- Table A6 (Manufacture of Plastic Components for Construction, industry 252E): Labor Productivity medians: Non-Cartel Members 4.758, Cartel Members 5.585; Log Sales medians: Non-Cartel Members 7.695, Cartel Members 10.516.
- Table A7 (Estimation of κ, Inverse Markup, # Observations 2,235):
  - Firm’s Market Share coefficient examples: -0.531*** (0.176) in column (1); Cartel’s Market Share -0.320*** (0.052) in column (1).
  - Intercept 0.704*** (0.009).
  - Implied κ 0.70 (column 1).
- Table A8 (Price-Fixing Cartels, # Observations 931):
  - Firm’s Market Share -0.682*** (0.188) in column (1); Cartel’s Market Share -0.320*** (0.162) in column (1).
  - Intercept 0.684*** (0.013). Implied κ 1.42 (column 1).
- Table A9 (Markup Distribution): Unconditional p50 1.098, p75 1.099, p90 1.102, p95 1.110, p99 1.341; Sectoral benchmark p50 1.177, p75 1.221, p90 1.307, p95 1.381, p99 1.674; SD log unconditional 0.032, competitive economy SD log 0.014.

### C. Figures and key empirical patterns
- Figure A1: Number of Decisions per Year (visual series 1995–2017).
- Figure A2: Number of Anti-competitive Firms per Year (visual series 1994–2007).
- Figure A6 (Impact of Collusion on Sectoral Productivity): plots p25, median, p75 of sectoral productivity growth for different κ values; example statement: "For instance, for κ=1, 75% of cartelized sectors have a productivity growth rate lower than -1.5%."
- Figure A7 (Ball Bearing industry price-fixing 1993-2003): visual shows price index and price change with black vertical lines at instances of concerted price increases; red vertical line = date when Competition Authority gave its sentence; dark gray area = +/- 2 standard deviations of mean monthly price change.

### D. Identification and estimation (key analytical points and parameter identification)
- Restriction on parameters (from equation (24) and related reasoning):
  - Sum of two slopes from equation (24): a1 + a2 = 1/ρ − 1/η.
  - For non-cartel members, coefficients b0 = ρ−1/ρ and b1 := η−ρ/(ηρ) satisfy b1 = 1/ρ − 1/η = a1 + a2.
  - η can be expressed as a function of ratio b1/b0 and ρ: η = (1/ρ − b1/b0 ((ρ−1)/ρ))−1.
  - Authors choose to rely on equation (24) projecting cartel members’ markup on own market share and cartel’s market share.
- Parameter identification via Jacobian (Figure A8) — moments’ sensitivities to parameters (percent change in each moment following a one percentage point increase in each parameter) — qualitative implications:
  - (i) Increase in ρ increases relative demand for more productive firms and increases aggregate markup.
  - (ii) Ratio of estimated coefficients from equation (24) helps identify η; increase in η increases this ratio and strengthens positive relationship between market share and markups.
  - (iii) Pareto shape parameter ξ: when ξ increases, productivity distribution is less skewed → changes sales concentration moments.
  - (iv) Geometric parameter σ identified by matching median number of firms per sector: higher σ → fewer firms → affects sales concentration and aggregate markup.
  - (v) Geometric parameter ζ identified by matching median number of cartel members: lower ζ → higher median number of cartel members → affects sales concentration.

### E. Mathematical Appendix — model structure, equilibrium, and welfare (key equations and findings preserved)
- Cartel objective with cross-ownership and profit weights:
  - Weighted portfolio profit objective: ˜πk = ∑l γkl ∑j βjl πj (equation (29)), reduce to ˜πk ∝ πk + ∑j∈C\{k} κkj πj (equation (30)) where κkj := ∑l γkl βjl / ∑l γkl βkl.
  - Cartel member maximization (equation (31)):
    - max over Psk, qsk of [(Psk − W zsk) qsk + ∑j∈C\{k} κkj (Psj − W zsj) qsj], ∀ k ∈ Cs.
- Inverse demand and markups:
  - Combined inverse demand: (Psk/P) = (qsk/ys)−1/ρ (ys/c)−1/η (equation (32)).
  - Equilibrium prices: P̃sk = μ̃sk W z k for k ∈ C, Psi = μsi W z i for i ∉ C (equation (35)).
  - Firm-level markups:
    - 1/μ̃sk = ρ−1/ρ + (η−ρ)/(ηρ) (ωsk + ∑j∈C\{k} κkj ωsj), ∀ k ∈ C.
    - 1/μsi = ρ−1/ρ + (η−ρ)/(ηρ) ωsi, ∀ i ∉ C. (equation (36))
  - Market share definition: ωsk := (Psk qsk) / ∑K j=1 (Psj qsj) = (Psk/Ps)1−ρ (equation (37)).
- Special collusion cases (exact formulas preserved):
  - Full symmetric collusion (κkj = 1) → 1/μ̃sk = ρ−1/ρ + (η−ρ)/(ηρ) ∑j∈C ωsj (equation (41)).
  - Partial symmetric collusion (constant κ across cartel members) → 1/μ̃sk = ρ−1/ρ + (η−ρ)/(ηρ) (ωsk + κ ∑j∈C\{k} ωsj) (equation (42)), κ ∈ (0, 1).
- Proposition 1 (summary of proof results and comparative statics):
  - i) Symmetric collusion increases the sectoral price index.
  - ii) Symmetric collusion increases markups of all firms.
  - iii) For cartel members, the markup increase declines with firm size (derivation uses Υsk ωsk decreasing with size if and only if ρ > 2; quantitative analysis uses ρ > 2).
  - iv) For non-cartel members, the markup increase increases with firm size.
  - Key intermediary expressions:
    - Umbrella pricing effect: Υsk := ωsk (ρ−1) (1/η − 1/ρ) μsk / [1 + ωsk (ρ−1) (1/η − 1/ρ) μsk] (equation (48)).
    - Sectoral price change under small ∆κ (equation (56)):
      - ˆPs = [1 / (1 − ∑k ωsk Υsk)] × [1/(ρ−1) ∑k∈C Υsk (ωsC − ωsk)] ∆κ.
- Collusion and misallocation:
  - Sectoral productivity change expression (equation (64)):
    - ˆzs = ∑k ωsk (μs/μsk − 1) ˆPsk + (ρ−1) ∑k ωsk μs/μsk (ˆPsk − ˆPs).
- Alternative collusion arrangements and overcharges:
  - General inverse markup with firm-specific κk (equation (65)):
    - 1/μ̃sk = ρ−1/ρ + (1/η − 1/ρ) (ωsk + κk ∑j∈C\{k} ωsj).
  - Overcharge for firm k joining cartel (equation (68)):
    - Θsk = [1/(ρ−1)] Υsk ωsk (ωsC − ωsk) ∆κk.
  - Sectoral price change is weighted average of overcharges times multiplier (equation (69)).
  - A class of collusion arrangements parametrized by ∆κk = ψ(ωsk) ψC ∆κ produces identical sectoral price change to a symmetric cartel if ψ chosen as in equation (71).
- Cartel stability and incentive constraints:
  - Change in log profits after cartel formation (no tech shocks): ˆΠCsk = μsk/(μsk − 1) ˆμsk + ˆqsk (equation (74)).
  - Using demand relations, change in profits expressed (equation (77)):
    - ˆΠCsk = [ρ − η − Υsk (ρ − εsk)] ˆPCs − (ρ − εsk) Θsk.
  - Participation constraint for non-cartel members: ˆΠsk = [ρ − η − Υsk (ρ − εsk)] ˆPCs > 0 (equation (78)).
  - Participation constraint for cartel members requires Θsk < [ρ − η)/(ρ − εsk) − Υsk] ˆPCs (equation (79)).
  - Incentive compatibility for trigger strategy: existence of δ such that (1/(1 − δ)) ˆΠCsk > ˆΠC\{k}sk (equation (80)); detailed condition in equation (83).
- Bertrand competition variant:
  - Under Bertrand, first-order yields Psk = [ρ − (ρ − η) ωsk] / [ρ − (ρ − η) ωsk − 1] W zsk (equation (86)).
  - Demand elasticities: ε(ωsk) = ρ − (ρ − η) ωsk (equation (87)).
  - Cartel member demand elasticity becomes ε(s) = ρ − (ρ − η) (ωsk + ∑j∈C\{k} κkj ωsj) (equation (88)).
  - Qualitative effects similar to Cournot but markups typically smaller under Bertrand.
- Consumption-equivalent welfare change (exact formulas preserved):
  - Lifetime utility: W = ∑t=0^∞ β^t (ln Ct − L_t^(1+ψ)/(1+ψ)) (equation (89)).
  - Consumption-equivalent welfare change ∆C solves (equation (91)):
    - ∆C = exp((WCartels − WCompetitive) (1 − β)) − 1.
  - Interpretation: ∆C accounts for cost of transition to the competitive Nash-Cournot steady-state.

*Source: Macroeconomic Effects of Market Structure Distortions: Evidence from French Cartels — Working Paper No. WP/2022/104*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022104-print-pdf.pdf_