## wpiea2022007-print-pdf

## Source details

**Canonical URL:** [wpiea2022007-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022007-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022007-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022007-print-pdf.pdf.json)

---

### Key empirical findings on distribution costs
- Aggregate magnitude
  - Manufacturing plants spent an average of 3.2% of their sales on distribution between 1993 and 2013.
  - Distribution costs are more than half as large as labor costs, which averaged 6.0% of aggregate sales over the same period.
  - Relative to value-added, the aggregate distribution share is 10.3% and the labor share 19.1%.
  - The manufacturing sector’s average intermediate share of sales is 77%.

- Heterogeneity across firms
  - Distribution shares increase with plant size: smallest decile ≈ 1% of sales; largest decile ≈ 3.5% of sales.
  - Summary statistics (pooling 1993–2013): median distribution share = 0.7%; mean = 2.1%; 90th percentile = 5.6%; share of plants reporting zero distribution expenses = 25.1%.
  - As a share of value-added: median = 3.0%, mean = 8.4%, 90th percentile = 23.7%.

- Time trends (late 1990s to late 2000s)
  - Aggregate distribution share declined from around 3.8% to around 2.5% of sales between 2000 and 2010.
  - Decline particularly pronounced for larger firms.
  - Ratio of aggregate to average distribution costs fell from 1.6 to 1.4 between 2000 and 2010.
  - Interstate trade share increased from 23% to 32% between 2000 and 2010.

### Data and measurement
- Data source and coverage
  - Indian Annual Survey of Industries (ASI), 1993–2013; on average over 30,000 plants per year.
  - ASI records plant-level “outward” distribution expenses (transportation costs, rebates, commissions, transit insurance of goods sold, packing fees for goods sold); excludes fixed distribution costs such as advertising or setting up a warehouse.
  - Distribution costs reported as a share of sales (sales net of taxes).

- Measurement issues addressed
  - Inward distribution costs (transport and acquisition of production materials) are decreasing in plant size, countering measurement-error explanations.
  - Pattern holds when excluding exporters.
  - Distribution share relative to costs is increasing in plant size, ruling out explanation via decreasing profit shares with plant size.

### Model structure and identification
- Model overview
  - Closed-economy, continuous-location general equilibrium model built on Melitz (2003) and Arkolakis (2010), with multiple sectors and intermediate inputs.
  - Manufacturing firms differ in variety quality (ψ) and choose which locations to serve.
  - Selling to a location requires setting up a local distributor (fixed cost increasing with distance); shipping incurs variable cost increasing with distance.
  - Larger, more productive firms sell to more distant locations; distribution expenses act like an increasing marginal cost.

- Key parameters governing distribution in calibration
  - (i) Productivity of the distribution sector (A_D).
  - (ii) Rate at which variable distribution costs rise with distance (ε).
  - (iii) Rate at which fixed costs of accessing markets rise with distance (β).

- Identification strategy
  - Aggregate distribution share primarily identifies A_D.
  - Rate at which variable costs rise with distance (ε) identified by how spending of large firms responds (ratio of aggregate to average distribution share).
  - Interstate trade share primarily identifies how fixed costs of shipping rise with distance (β).

### Quantitative findings and welfare analysis
- Calibration targets and baseline
  - Model calibrated to year 2000 using method of simulated moments to match distribution share facts, firm size distribution, and estimated interstate trade; re-calibrated to 2010 holding model structure constant except for A_D, ε, β.

- Baseline (2000) calibrated parameter values (selected)
  - θ = 0.77 (intermediate share) — Data/Model match: 77%.
  - δ = 0.082 (exit rate) — Data/Model match: 8.2%.
  - A_D = 3.3 — aggregate distribution share matched: Data 3.8% — Model 3.8%.
  - ε = 38.5 — aggr / avg distribution share matched: Data 1.62 — Model 1.62.
  - β = 47.7 — share of sales across states matched: Data 23.0% — Model 23.0%.
  - η = 4.1 — sales share top 10%: Data 83.0% — Model 85.9%.
  - f_e = 0.2 — sales share bottom 50%: Data 1.7% — Model 1.7%.
  - f = 3.5 — moment: mass of firms.

- Counterfactuals and welfare impacts (selected)
  - No distribution costs (A_D = ∞): welfare gain +47.2%.
  - Eliminating variable distribution requirements raises interstate trade share by 11%.
  - Employment: top 10% employment share increases by 0.7%; bottom 50% employment share nearly 5% lower under no-distribution-costs counterfactual; exit threshold increases by 2%.
  - Setting ε = 0 (distribution costs constant with distance): welfare gain +16.7%.
  - A 10% reduction in A_D:
    - with ε = 38.5 → welfare change −3.0%;
    - with ε = 0 → welfare change −2.1%.
  - Thicker tail of firm productivity (η ↓ 1%):
    - with ε = 38.5 → +14.0%;
    - with ε = 0 → +15.9%.
  - Amplification: with calibrated ε, a 10% decline in A_D leads to a 40% amplification of welfare losses relative to ε = 0 case.

### Changes 2000 → 2010 and inferred parameter shifts
- Observed moment changes 2000 → 2010
  - Aggregate distribution share: 3.8% → 2.4% (Table 4 moment: 2.4%).
  - Aggregate / average distribution share: 1.6 → 1.39.
  - Interstate trade share: 23% → 31.7% (reported 32% elsewhere; calibration moment 31.7%).

- 2010 calibrated parameter values (selected)
  - A_D = 5.9 — moment: aggregate distribution share — Data 2.4% — Model 2.4%.
  - ε = 18.0 — moment: aggregate / average distribution share — Data 1.39 — Model 1.39.
  - β = 38.2 — moment: share of sales across states — Data 31.7% — Model 31.7%.

- Interpretation
  - Increase in A_D and decrease in ε imply higher distribution-sector productivity and a flattening of how variable costs rise with distance.
  - Decline in β implies lower fixed costs of accessing distant markets, consistent with increased interstate trade.
  - Suggested driver: large transport infrastructure investments in the 2000s (e.g., National Highways Development Project; Golden Quadrilateral launched 2001, >90% finished by 2006, completed 2012).

- Welfare impacts 2000 → 2010
  - Combined improvements in A_D, ε and β imply welfare gains of 58.2% over the decade.
  - Decomposition of welfare gains:
    - All three parameters → 58.2%.
    - A_D only → 14.4%.
    - ε only → 7.6%.
    - β only → 25.2%.
  - Sum of individual impacts ≈ 47.2% (sum of listed components) — less than total 58.2% due to complementarities across parameters.

### Empirical robustness and additional empirical patterns
- Distribution cost decomposition (ASI 1993–1997)
  - Transport costs: 46.4% of distribution expenses.
  - Commissions: 17.8%.
  - Rebates: 12.1%.
  - Other: 23.7%.
  - All components behave similarly as a function of plant size; total spending used in analysis.

- Evidence on robustness
  - Relationship between distribution share and plant size holds when:
    - using different industry classifications and fixed effects,
    - restricting to non-exporters,
    - measuring distribution share relative to total costs.
  - Inward distribution share is decreasing in plant size (years 1993, 1994, 1996), countering mismeasurement concerns.

- Figures and data notes (selected)
  - Histogram of distribution share of sales: ASI 1993–2013, winsorized at 20%.
  - Binned scatter plots show distribution share increasing in plant size, residualized on sector×year fixed effects.
  - Distribution share of total costs vs plant size also increasing; total costs include intermediate expenditures, labor costs and imputed rental on capital assuming rental rate = 20%.

### Economic interpretation and mechanisms
- Core channel
  - Variable distribution services per unit increase with distance (ε > 0), so marginal costs rise with distance.
  - Higher-quality (more productive) firms sell farther (larger n*(ψ)), thus incur larger distribution shares.
  - Low productivity in distribution disproportionately constrains the most productive firms’ geographic scope, allowing smaller, inefficient producers to survive and lowering aggregate output and welfare.

- Model comparative statics and equilibrium effects
  - Decrease in ε expands firms’ market reach n*(ψ), raises wages via general equilibrium, increases production costs and raises the productivity cutoff ψ̄ for active firms — resulting equilibrium has larger and higher-quality firms.
  - Parameters β and ε both reduce market reach when higher, but affect distribution shares differently; this distinction used for identification.

### Policy-relevant findings and recommendations
- Empirical and model-based conclusions
  - Distribution expenses are large in aggregate and disproportionately borne by larger firms.
  - Size-dependent distribution frictions amplify aggregate welfare losses from low TFP in distribution by 1.5 times relative to a world with equal distribution shares.
  - Distribution friction is an important bottleneck when the intermediate share θ = 0.77 is high.

- Policy implications
  - Investments in transport infrastructure (e.g., road networks) can yield large welfare gains, but maximal benefits require complementary policies that lower fixed market-access costs (e.g., reducing entry barriers for distributors).
  - Because TFP in distribution disproportionately affects larger firms, high distribution costs reduce private incentives for firms to innovate or invest; lowering domestic market-expansion costs increases private returns to such investments.

*Source: wpiea2022007-print-pdf - Introduction, Appendix figures, Calibration, Quantitative analysis, Data Appendix, and References.*

### Introduction ...........................................................................................................

### Introduction

### Key empirical findings on distribution costs
- Aggregate magnitude
  - Manufacturing plants spent an average of 3.2% of their sales on distribution between 1993 and 2013.
  - Distribution costs are more than half as large as labor costs, which averaged 6.0% of aggregate sales over the same period.
  - Relative to value-added, the aggregate distribution share is 10.3% and the labor share 19.1%.
  - The manufacturing sector’s average intermediate share of sales is 77%.

- Heterogeneity across firms
  - Distribution shares increase with plant size: the smallest decile of plants spend close to 1% of sales on distribution, while the largest decile spend around 3.5% of sales.
  - Summary statistics (pooling 1993–2013): median distribution share = 0.7%; mean = 2.1%; 90th percentile = 5.6%; share of plants reporting zero distribution expenses = 25.1%.
  - As a share of value-added, median = 3.0%, mean = 8.4%, 90th percentile = 23.7%.

- Time trends (late 1990s to late 2000s)
  - The aggregate distribution share declined from around 3.8% to around 2.5% of sales.
  - The decline was particularly pronounced for larger firms.
  - Ratio of aggregate to average distribution costs fell from 1.6 to 1.4 between 2000 and 2010.
  - Interstate trade share increased from 23% to 32% between 2000 and 2010.

### Data and measurement
- Data source and coverage
  - Data come from the 1993–2013 waves of the Indian Annual Survey of Industries (ASI).
  - The ASI is a nationally representative survey of formal manufacturing plants; on average over 30,000 plants per year.
  - ASI records plant-level “outward” distribution expenses (transportation costs, rebates, commissions, transit insurance of goods sold, packing fees for goods sold). This captures variable costs of selling goods but excludes fixed distribution costs such as advertising, finding distributors, or setting up a warehouse.
  - Distribution costs are reported in the paper as a share of sales (sales net of taxes).

- Measurement issues addressed
  - Evidence against measurement error driving size patterns: inward distribution costs (costs to transport and acquire production materials) are decreasing in plant size.
  - Pattern holds when excluding exporters.
  - Distribution share relative to costs is also increasing in plant size, ruling out explanation via decreasing profit shares with plant size.

### Model structure and identification
- Model overview
  - A closed-economy, continuous-location general equilibrium model built on Melitz (2003) and Arkolakis (2010), with multiple sectors and intermediate inputs.
  - Manufacturing firms differ in variety quality (productivity) and choose how many and which locations to serve.
  - Selling to a location requires setting up a local distributor (captured as a fixed cost that can increase with distance).
  - Shipping across space incurs a variable cost that can increase with distance.
  - Larger, more productive firms sell to more distant locations; distribution expenses act like an increasing marginal cost, disproportionately affecting the most productive firms.

- Key parameters governing distribution in the calibration
  - (i) Productivity of the distribution sector.
  - (ii) Rate at which variable distribution costs rise with distance.
  - (iii) Rate at which fixed costs of accessing markets rise with distance.

- Identification strategy
  - Aggregate distribution share primarily identifies distribution-sector productivity (through the price of distribution services).
  - The rate at which variable costs rise with distance is identified by how spending of large firms responds (ratio of aggregate to average distribution share).
  - The interstate trade share primarily identifies how fixed costs of shipping rise with distance.

### Quantitative findings and welfare analysis
- Calibration targets and periods
  - Model calibrated to year 2000 using method of simulated moments, matching distribution share facts, firm size distribution, and estimated interstate trade.
  - Recalibrated to 2010 to evaluate changes over 2000–2010.

- Counterfactuals and welfare impacts
  - Eliminating variable distribution requirements (i.e., an economy without these distribution inputs) would raise welfare by 47%.
  - Firm heterogeneity matters: an economy in which all firms use distribution inputs equally would have welfare 17% higher than the baseline (holding other things constant).
  - Heterogeneity amplifies losses from reductions in distribution-sector TFP: aggregate losses from a reduction in distribution sector TFP are amplified 1.5-fold relative to an economy where all firms spend equally on distribution inputs.

- Gains from 2000–2010 changes
  - From 2000 to 2010: aggregate distribution share fell from 3.8% to 2.5%; ratio of aggregate to average distribution costs fell from 1.6 to 1.4; interstate trade share rose from 23% to 32%.
  - Improvements inferred in the three distribution parameters jointly imply total welfare gains of 58% over the ten years.
  - Decomposition of contributors (individual impacts):
    - Reduction in fixed cost of accessing new markets: welfare gains of 25%.
    - Increase in distribution-sector TFP: welfare gains of 14%.
    - Reductions in the rate at which distribution costs increase with distance: welfare gains of 8%.
  - The sum of individual impacts is 47%, smaller than the total 58% due to complementarities between parameters (e.g., distribution-sector TFP improvements are less effective with large fixed access costs).

### Economic interpretation and mechanisms
- Channel emphasized
  - Larger, more productive firms are more exposed to distribution-sector inefficiencies because they ship further and incur larger distribution shares, so low productivity in distribution disproportionately constrains firm size and the geographical scope of sales.
  - Restriction on large productive firms allows smaller, inefficient producers to survive, reducing aggregate output and welfare.
  - Firm-level heterogeneity in exposure to an upstream sector (distribution) amplifies intersectoral complementarities and aggregate losses from low productivity in that sector.

### Relation to literature
- Complements intranational trade literature that infers internal trade barriers from price variation and flows by providing direct plant-level measures of distribution expenses.
- Extends literature on intersectoral linkages and development by documenting firm heterogeneity in exposure to distribution inputs and quantifying its aggregate importance.
- Connects to literature on firm size distributions in developing vs. developed economies via a mechanism where distribution costs increase with plant productivity and thus help explain mass of small firms.

*Source: wpiea2022007-print-pdf - Introduction*

### Appendix Figure5 that the relationship between the distribution share and size holds

### Appendix Figure5 that the relationship between the distribution share and size holds

### Decomposition of distribution costs (ASI 1993–1997)
- Distribution costs decomposed into four sub-components for 1993-1997:
  - (1) transport costs (freight and shipping, packing fees, transit insurance)
  - (2) commissions owed to intermediaries
  - (3) rebates and drawbacks owed to customers
  - (4) other
- Aggregate shares of distribution expenses:
  - Transport costs: 46.4%
  - Commissions: 17.8%
  - Rebates: 12.1%
  - Other: 23.7%
- Empirical finding: all components of distribution expenses (as a share of sales) behave similarly as a function of size; therefore total spending on distribution is used for the remainder of the analysis.

### Empirical patterns: Distribution share vs plant size
- Data source: Indian Annual Survey of Industries 1993 to 2013.
- Figure 1 notes:
  - Binned scatter plots of plant-level distribution share against log(output), residualized on sector-year fixed effects.
  - Left panel pools all years; right panel shows 1993-1997 and plots four subcomponents (transport, commissions, rebates, other), residualized on sector×year fixed effects.
- Main empirical pattern: distribution costs are large and increasing in plant size.

### Trends in the distribution share over time
- Aggregate trend (Figure 2(a)):
  - Aggregate distribution costs relative to sales fell considerably from 2000 to 2010.
  - Distribution share fell from just below 4% to around 2.5% during this period.
  - Period coincided with large transport infrastructure investments (see footnote 9).
- Plant-level heterogeneity (Figure 2(b)):
  - The decline in the aggregate distribution share was driven by changes in the distribution share conditional on plant size (not by changes in the size distribution of plants).
  - The distribution share of sales decreased for plants of all sizes between the compared windows (1999-2001 vs 2009-2011).
  - Sales share movements: the smallest 50% of plants: 1.7% → 1.3%; the largest 10% of plants: 83.3% → 84.5% (footnote 10).
  - Large plants experienced even larger decreases in their distribution shares than smaller plants, producing a flattening in the relationship between distribution share and plant size.
- Implication: reductions in distribution costs yield important potential gains, particularly for larger manufacturing plants.

### Robustness and alternative explanations
- Proposed primary explanation:
  - (i) variable cost of shipping goods increases with distance shipped; and
  - (ii) larger plants tend to ship further distances than smaller plants.
  - Supported by trade literature documenting transport costs increasing in distance and evidence that larger plants ship further (Holmes and Stevens (2012); Bernard and Jensen (1995) referenced).
- Addressed alternative explanations:
  1. Systematic mismeasurement (smaller plants underreporting distribution costs or reporting in labor):
     - Counterevidence: inward distribution share is decreasing in plant size (Appendix Figure7), mitigating mismeasurement concerns.
     - Inward distribution expenses reported in ASI for years 1993, 1994 and 1996 (footnote 12).
  2. Larger plants charging higher markups (lower cost shares overall):
     - Counterevidence: Appendix Figure8 shows the distribution share of total costs is also increasing in plant size; the decrease in aggregate distribution share in Figure 2(a) is similar when measured relative to total costs.
  3. Exporters driving the relationship:
     - Counterevidence: Appendix Figure6 shows the positive relationship between distribution share and size holds when sample is restricted to non-exporters.
     - Exports are reported in ASI for years 1997, 2008, 2009, 2011, 2012 and 2013 (footnote 13).

### Conceptual summary and implications
- Empirical summary:
  - Aggregate distribution share in Indian manufacturing is large.
  - Bigger firms spend more on distributing their goods (distribution share increases with plant size).
  - The increasing relationship arises from transport costs increasing in distance and larger firms shipping farther, making larger firms more intensive users of distribution services.
- Policy-relevant implication:
  - Large firms are particularly affected by the quality of Indian infrastructure and the productivity of the distribution sector.
  - The pronounced decline in aggregate distribution share during the 2000s was especially notable for the largest firms.

### Model overview and key mechanisms (sections 3.1–3.3)
- Model elements:
  - Closed-economy, continuous-location extension of Arkolakis (2010); economy lies on a circle with circumference 1.
  - Two sectors: manufacturing and distribution.
  - Final goods producers at each location combine varieties via CES aggregator with elasticity σ; varieties differ in quality ψ.
  - Distribution sector: perfectly competitive, constant returns to scale d = A_D L_D, price p_D = w/A_D.
  - Manufacturing firms heterogeneous in quality ψ and location; choose markets to serve and quantities. Firms face:
    - Fixed cost f(n) = f(1−2n)^(−β) of accessing a market at distance n (fixed costs in units of labor).
    - Distribution services requirement d(n) = 1 + ε n units of distribution services per unit sold to distance n.
    - Leontief constraint: q(n) = min{ M^θ L_p^(1−θ), d(n)^(1+ϵ n) } (stated as fixed proportions; elasticity remarks present).
  - Firm profit maximization yields per-market profits π*(ψ,n) and optimal furthest market n*(ψ) determined by maximizing integrated profits up to n*(ψ).
  - Entry: sunk entry cost f_E; qualities drawn from distribution g(ψ); exit probability δ each period; firm value = π*(ψ)/δ.
- Key mechanism generating empirical facts:
  - With ε > 0, marginal distribution services per unit increase with distance, so marginal costs increase in n; higher-quality firms sell farther (larger n*(ψ)), hence their sales include a larger share of sales devoted to distribution.
  - Derived expression (interior solutions) for n*(ψ):
    - n*(ψ) = 1/ε ( A_D c (σ−1)/σ (L / f σ)^{1/(σ−1)} ψ^{σ/(σ−1)} − (A_D + 1) )
    - where c = (1/θ)^θ (w^{1−θ})^{1−θ} (as defined in text).
  - Distribution share in market n:
    - p_D d(n,ψ) / (p(ψ,n) q(ψ,n)) = (1+ε n)/A_D c w + (1+ε n)/A_D (σ−1)/σ (stated as equation (13) in text).
  - Comparative statics:
    - Decrease in ε increases n*(ψ) for all firms (expansion), raises wage through general equilibrium, which increases production costs and raises the productivity cutoff ψ̄ for active firms, leading to an equilibrium with larger and higher-quality firms.
    - Parameters β and ε both make firms serve more locally when higher, but differ in their effects on the distribution share; this identification strategy is used for calibration.

### Quantitative analysis overview (section 4)
- Calibration targets: distribution shares and how they vary with plant size from ASI, plus newly constructed estimates of interstate trade.
- Key quantitative finding:
  - Improvements in the distribution sector between 2000 and 2010 (combining increases in distribution sector TFP A_D, reductions in fixed cost of accessing markets, and flattening of the distribution cost curve across space) led to cumulative welfare gains of around 58% over 10 years.

*Appendix Figure5 that the relationship between the distribution share and size holds — content from the provided PDF excerpt.*

### introduction of the Goods and Services Tax (GST) in 2017.

### introduction of the Goods and Services Tax (GST) in 2017.

### Data and measurement of interstate trade
- Raw data source: "Inter-State Movements / Flows of Goods by Rail, River and Air", published by the Indian Directorate General of Commercial Intelligence and Statistics (DGCIS).
- Two limitations of DGCIS data:
  - shipments measured in quantities rather than values;
  - exclude shipments by road.
- Estimation approach (based on Van Leemput (2021)):
  - Uses that interstate shipments for export are reported separately from interstate shipments for domestic consumption.
  - Assumptions required:
    - 1) for each good, the share of shipments by rail vs road is the same for exports and domestic consumption;
    - 2) the quality of goods shipped interstate by rail is the same for export and for domestic consumption.
  - Implied identity (Equation (15)):
    - quantity exported via rail / quantity shipped interstate via rail = value exported via road and rail / value shipped interstate via road and rail
  - Interstate trade share for each good computed as: export intensity (aggregate export value / aggregate production value) divided by the ratio in Equation (15); averaged across goods to obtain targeted moments.
- Estimated interstate trade share:
  - increased from 23% in 2000 to 32% in 2010.
- Note: India’s Ministry of Finance estimated interstate trade in goods and services subject to the GST amounted to 60% of GDP in 2017 (Ministry of Finance, 2018).

### Calibration of the model (baseline year 2000)
- Model setup:
  - Firm productivity ~ Pareto with shape parameter η.
  - Total of 8 parameters; two set externally:
    - intermediate share θ = 0.77 to match 77% intermediate share measured in the ASI;
    - exit rate δ = 0.082 to match exit rates from Peters (2020) for formal manufacturing plants in Indonesia.
  - Remaining six parameters jointly calibrated to match key moments in 2000.
- Key parameters and identification notes:
  - A_D (productivity of distribution sector) primarily affects aggregate distribution share.
  - ε (rate at which distribution requirements rise with distance) strongly affects aggregate vs average distribution share; ε = 0 implies identical average and aggregate distribution shares.
  - β (rate at which fixed cost of accessing markets rises with distance) identified by share of sales across states; "sales in state" defined as sales within a radius corresponding to a typical state’s share of Indian GDP (~7% segment length considered within-state sales).
  - η and f_e calibrated to match employment distribution across firms.
- Table 2: Parameters and moments in 2000 (reported values)
  - (A) Set externally
    - θ 0.77 — moment: intermediate share — Data: 77% — Model: 77%
    - δ 0.082 — moment: exit rate (Peters(2020)) — Data: 8.2% — Model: 8.2%
  - (B) Jointly calibrated
    - A_D 3.3 — moment: aggregate distribution share — Data: 3.8% — Model: 3.8%
    - ε 38.5 — moment: aggr / avg distribution share — Data: 1.62 — Model: 1.62
    - β 47.7 — moment: share of sales across states — Data: 23.0% — Model: 23.0%
    - η 4.1 — moment: sales share top 10% — Data: 83.0% — Model: 85.9%
    - f_e 0.2 — moment: sales share bottom 50% — Data: 1.7% — Model: 1.7%
    - f 3.5 — moment: mass of firms — 1

### Welfare impacts: level and heterogeneity in distribution costs
- Empirical facts documented:
  - Indian manufacturing spends around 4% of sales on distributing their products (aggregate distribution share ~3.8% in 2000).
  - Bigger firms spend over three times more – as a share of sales – on distribution compared to small firms.
- Counterfactuals and main welfare results (Table 3 and textual results):
  - Level of distribution costs:
    - No distribution costs (A_D = ∞): welfare gain +47.2%.
    - Eliminating variable distribution costs raises interstate trade share by 11%.
    - Employment: employment share of top 10% of firms increases by 0.7%; bottom 50% employment share nearly 5% lower.
    - Exit threshold: increases by 2%.
  - Heterogeneity in distribution costs:
    - Setting ε = 0 (distribution costs constant wrt distance): welfare gain +16.7%.
    - A 10% reduction in A_D:
      - with ε = 38.5 → welfare change -3.0%;
      - with ε = 0 → welfare change -2.1%.
    - Thicker tail of firm productivity (η ↓ 1%):
      - with ε = 38.5 → +14.0%;
      - with ε = 0 → +15.9%.
  - Complementarities and amplification:
    - With calibrated ε, a 10% decline in A_D leads to a 40% amplification of welfare losses relative to ε = 0 case.
    - Size-dependent distribution costs limit gains from improvements in firm productivity; welfare gains from a 1% decrease in η are ~12% lower in baseline with calibrated ε vs ε = 0.

### Changes in the distribution sector over 2000–2010 and welfare implications
- Re-calibration to 2010: only A_D, ε, and β re-calibrated to match three moments: aggregate distribution share, ratio aggregate/average distribution share, interstate trade share.
- Key observed moment changes 2000 → 2010:
  - aggregate distribution share: fell from 3.8% to 2.4% (Table 4 moment: 2.4%).
  - aggregate / average distribution share: fell from 1.6 to 1.4 (Table 4 moment: 1.39).
  - interstate trade share: increased from 23% to 32%.
- Table 4: Parameters and moments in 2010 (reported values)
  - A_D 5.9 — moment: aggregate distribution share — Data: 2.4% — Model: 2.4%
  - ε 18.0 — moment: aggregate / average distribution share — Data: 1.39 — Model: 1.39
  - β 38.2 — moment: share of sales across states — Data: 31.7% — Model: 31.7%
- Interpretation of parameter changes:
  - Increase in A_D and decrease in ε imply higher distribution-sector productivity and a reduction in how variable costs rise with distance.
  - A decline in fixed costs of distribution (β) also required to match the increase in interstate trade.
- Potential driver: large investments in road network (National Highways Development Project; Golden Quadrilateral launched 2001, completed 2012, >90% finished by 2006) — effects mapped into increases in A_D, reduction in ε, and a decline in β.
- Welfare impacts of changes 2000 → 2010:
  - Combined improvements in distribution sector led to welfare gains of 58.2% (representative manufacturing consumer welfare in 2010 vs 2000).
- Decomposition of welfare gains (Table 5 reported values)
  - All three parameters → 58.2%
  - A_D only → 14.4%
  - ε only → 7.6%
  - β only → 25.2%
- Key insights from decomposition:
  - Reduction in β alone produced welfare gains >25%, driven by increased interstate trade.
  - Improvements in A_D and reductions in ε produced gains of 14.4% and 7.6% respectively.
  - Sum of individual contributions is only just over 80% of total gain, indicating complementarities across components (TFP, variable cost distance profile, fixed market-access costs).

### Policy-relevant findings and recommendations
- Empirical and model-based findings:
  - Distribution expenses are large in aggregate and disproportionately borne by larger firms.
  - Distribution friction acts as a bottleneck that amplifies production-chain costs when intermediate share is high (θ = 0.77).
  - Size-dependent distribution inefficiencies amplify aggregate welfare losses from low TFP in distribution by 1.5 times relative to a world with equal distribution shares.
- Policy implications:
  - Improvements in transport infrastructure (e.g., road networks) can yield large welfare gains, but maximal benefits require complementary policies that reduce fixed costs of market access (e.g., lowering entry barriers for distributors).
  - Because TFP in distribution disproportionately affects larger firms, high distribution costs reduce private incentives for firms to innovate or invest in cost-reducing measures; lowering domestic market-expansion costs can increase private returns to such investments.

*Source: wpiea2022007-print-pdf - introduction of the Goods and Services Tax (GST) in 2017.*

### References

### References and Data Appendix (wpiea2022007-print-pdf)

### References
- Bibliographic citations cited in the content unit include works on trade frictions, market penetration costs, trade logistics, infrastructure and economic development, intra-national trade costs, misallocation and productivity, exporters and wages, rail infrastructure impacts, input-output networks, and heterogeneous mark-ups. Representative entries:
  - Allen, Treb, “Information Frictions in Trade,” Econometrica, 2014, 82(6), 2041–2083.
  - Arkolakis, Costas, “Market Penetration Costs and the New Consumers Margin in International Trade,” Journal of Political Economy, 2010, 118(6), 1151–1199.
  - Arvis, Jean-Franois, Lauri Ojala, Christina Wiederer, Ben Shepherd, Anasuya Raj, Karlygash Dairabayeva, and Tuomas Kiiski, “Connecting to Compete 2018 : Trade Logistics in the Global Economy,” World Bank, 2018.
  - Donaldson, Dave, “Railroads of the Raj: Estimating the Impact of Transportation Infrastructure.,” American Economic Review, 2018, 108(4-5), 899–934.
  - Melitz, Mark, “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity.,” Econometrica, 2003, 71, 1695–1725.
  - OECD, “OECD Services Trade Restrictiveness Index: Policy trends up to 2020,” 2020.
  - Peters, Michael, “Heterogeneous Mark-Ups and Endogenous Misallocation,” Econometrica, 2020, 88(5).
- Several working papers and NBER papers are cited, including Atkin & Donaldson (July 2015, NBER Working Paper No. 21439), Bartelme & Gorodnichenko (2015, NBER Working Paper No. 21251), and others.
- The list includes sector- and country-specific sources such as Ministry of Finance, Economic Survey 2017-18, Chapter 2, “A New, Exciting Birds-Eye View of the Indian Economy Through the GST,” 2018.

### A. Data Appendix — A1. Annual Survey of Industries (ASI)
- Data source:
  - Indian Annual Survey of Industries (ASI) microdata for years 1993 to 2013, publicly available through India’s Ministry of Statistics and Programme Implementation (MOSPI).
  - Reference period: accounting year beginning 1st of April and ending 31st of March; surveys referenced by the earlier of the two years covered.
- Sampling and coverage:
  - Representative sample of plants with at least 10 workers (threshold is 20 workers for plants that don’t use power).
  - Two sampling schemes: Census (surveyed every year; plants with more than 100 workers and plants in 12 industrially ‘backwards’ states) and Sample (random within state × 3-digit industry).
  - Note: The size threshold for Census sample is 200 in some years during the 1990s.
  - Sampling weights are provided with the data.
- Variable construction:
  - Plant output = gross value of product sales + changes in inventories + other sources of revenue.
  - Employment includes paid and unpaid labor.
  - Labor costs include wages and salaries, bonuses, contributions to firms’ provident (pension) fund, and other welfare expenses.
  - Capital stock = average of beginning and end of year book value of the net fixed capital stock.
  - Intermediate inputs include materials and fuels consumed as well as other expenditures.
  - Harmonized sectoral classification: 50 manufacturing sectors consistently defined throughout the period.
  - Official NIC classification changes in 1998, 2004 and 2008; official concordances used to construct harmonized classification.
  - Output from ASI scaled up by 3/2 to account for informal production.

### A. Data Appendix — A2. Interstate Trade (DGCIS methodology and limitations)
- Data source:
  - Indian Directorate General of Commercial Intelligence and Statistics (DGCIS), “Inter-State Movements / Flows of Goods by Rail, River and Air”.
  - Data collected for tax purposes based on invoices; reported for 70 goods categories (agricultural and manufactured goods); shipments between all Indian states reported, with separate entries for shipments for international export.
  - Exports reported for main port states of Andhra, Gujarat, Karnataka, Kerala, Maharashtra, Tamil Nadu and West Bengal. Delhi excluded in reporting; Delhi accounts for 8% of Indian exports (Van Leemput, 2021).
  - Van Leemput (2021) used this data to calculate India’s interstate trade share for 2011-2012.
- Approach to constructing interstate trade share (adjusted for shipments by road):
  - For each commodity, construct:
    - quantity exported by rail
    - quantity shipped interstate by rail
    - (Equation 16 in source)
  - Assumptions:
    1) For each commodity, the share of shipments by rail is the same for exports and domestic consumption.
    2) The quality of goods shipped interstate by rail is the same for export and for domestic consumption.
  - Under these assumptions:
    - quantity exported by rail / quantity shipped interstate by rail = aggregate value exported / aggregate value shipped interstate (Equation 17)
  - Export intensity for each commodity:
    - export intensity ≡ aggregate value exported / aggregate value of domestic production (Equation 18)
  - Interstate trade share for each commodity obtained by dividing Equation 19 by 17:
    - interstate trade share = aggregate value exported / aggregate value of domestic production × aggregate value exported / aggregate value shipped interstate (Equation 19)
- Practical implementation limitations:
  - Can only implement for commodities exported in both 2000 and 2010 because India’s trade liberalization was still underway in 2000; restricting to commodities with positive international exports reduces sample to 16 goods.
  - Need comparable data on aggregate export values and aggregate production for each commodity; data sourced from ASI, Indiastat and CEIC.
  - DGCIS commodity classification concorded manually to other classifications.
  - Some interstate trade shares excluded when >100% or when change between 2000 and 2010 is unrealistically large (e.g., ‘Other grains’ increases from 1.1% in 2000 to 96.5% in 2010).
  - After concordance and measurement checks, final set used: 5 categories — coal and coke, metal products, pulses, sugar and molasses, and iron and steel.
  - Data moments: average interstate trade share across these 5 categories in each year.
  - Note: The main state excluded (Delhi) is omitted in both 2000 and 2010; authors state they don’t expect this to bias estimates of the change in the interstate trade share.

### A. Data Appendix — A3. Figures (descriptions and data notes)
- Figure 4 — Histogram of Distribution Share of Sales:
  - Source: Indian Annual Survey of Industries 1993 to 2013.
  - Note: Histogram of distribution share of sales in India, winsorized at 20%.
- Figure 5 — Distribution Share vs Plant Size: Robustness:
  - Source: Indian Annual Survey of Industries 1993 to 2013.
  - Both subfigures show binned scatter plots of plant-level distribution share against plant size, residualized on industry×year×state fixed effects.
  - Industry classifications used: 4-digit NIC87, 5-digit NIC98, 5-digit NIC04 and 5-digit NIC08 (between 400 and 700 manufacturing industries at this level).
  - Subfigure (a) uses output as plant size; subfigure (b) uses labor.
- Figure 6 — Distribution Share of Sales for Non-Exporters:
  - Source years: Indian Annual Survey of Industries 1997, 2008, 2009, 2011, 2012 and 2013.
  - Binned scatter plot of plant-level distribution share against output for plants reporting no exports; both variables residualized on sector×year fixed effects.
- Figure 7 — Inward Distribution Share vs. Plant Size:
  - Source years: Indian Annual Survey of Industries 1993, 1994 and 1996.
  - Binned scatter plot of plants’ inward distribution share of materials and fuels against log(output), residualized on sector×year fixed effects.
  - These years are the only ASI years reporting inward distribution costs (expenses for acquiring and transporting production materials).
- Figure 8 — Distribution Share of Total Costs vs. Plant Size:
  - Source: Indian Annual Survey of Industries 1993-2013.
  - Binned scatter plot of plant-level distribution share of total costs against output, residualized on sector×year fixed effects.
  - Total costs include intermediate expenditures, labor costs and the imputed rental on capital assuming a rental rate of 20%.

### B. Model Appendix — Identification and Sensitivity Figures
- Figure 9 — Identification 2000: all parameters:
  - Shows sensitivity of model moments to parameters.
  - Starting from baseline calibration, each panel plots all six moments relative to their baseline value as a function of one parameter, keeping other parameters at baseline.
  - The change in each parameter is measured relative to baseline.
- Figure 10 — Identification 2010:
  - Shows sensitivity of model moments to parameters for 2010 calibration.
  - Starting from baseline, each panel plots all three moments relative to baseline as a function of one parameter, holding others constant.
  - The change in each parameter is measured relative to baseline.

*Content extracted from wpiea2022007-print-pdf (References and Data/Model Appendices).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022007-print-pdf.pdf_