## 15. Robustness for Export Volumes

## Source details

**Canonical URL:** [15. Robustness for Export Volumes](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1442.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1442.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1442.pdf.json)

---

### Introduction and motivation
- Empirical background: pass-through of exchange rate changes to domestic/import prices is typically incomplete; incomplete pass-through implies deviations from the Law of One Price and has implications for monetary and exchange rate policy.
- Research gap: scarce evidence on the role of product-level quality in explaining heterogeneous exchange rate pass-through.
- Strategy: focus on the wine industry where “quality” is observable via expert ratings; combine Argentinean firm-level destination-specific export values and volumes with Wine Spectator and Robert Parker ratings.

### Data and sample
- Customs trade data (Argentina): export flows 2002–2009 with exporter name, destination, date, 12-digit HS code, FOB value (US dollars), and volume (liters); product detail includes wine name, grape, type, vintage year.
- Final cleaned sample:
  - 21,647 different products/wines overall.
  - 6,720 wines matched with quality rankings (31 percent of wines, representing 58 percent of total FOB value exported between 2002 and 2009).
- Firms and scope:
  - 6,720 wines exported by 209 wine producers.
  - Exporters exported an average of 139 different wines (range 1 to 510).
  - Firms exported to an average of 40 different destinations (range 1 to 88).
- Quality measures:
  - Wine Spectator scores on a (50,100) scale: mean ranking 85; min 55; max 97.
  - Parker rankings (sample: 3,969 wines) on a (50,100) scale: mean 87; min 72; max 98.
  - Overlap: 2,433 wines have both rankings; correlation between the two rankings is 0.53.
- Macroeconomic data:
  - GDP from Penn World Tables.
  - CPI and nominal exchange rates from IFS.
  - Real exchange rate defined as ratio of CPIs times average yearly nominal exchange rate (an increase captures a real depreciation).
  - Argentina experienced a peso depreciation of up to 75 percent in 2002 after the 2001 financial crisis; further depreciation in 2008.

### Theoretical framework and testable predictions
- Model extension: Corsetti and Dedola (2005) extended to multi-product firms heterogeneous in product quality; local additive distribution costs paid in importing-country currency imply distribution costs unaffected by exchange rate changes.
- Key model implications:
  - Demand elasticity perceived by firms falls with a real depreciation and with quality.
  - Higher quality goods have higher marginal costs and higher markups.
- Predictions:
  - Prediction 1: elasticity of export price p_j(φ) to a change in real exchange rate q_j increases with product quality s(φ).
  - Prediction 2: elasticity of export volumes x_j(φ) to a change in q_j decreases with product quality s(φ).
  - Predictions 3 & 4: effects in Predictions 1 and 2 are stronger for high income destination countries (preference for quality increases with per capita income).

### Empirical framework
- Price (unit value) specification (reduced-form):
  - lnUV_kij,t = β1 ln q_j,t + β2 s_k^WS + β3 (ln q_j,t × s_k^WS) + controls + fixed effects + error.
  - UV = export unit value in pesos per liter; q_j,t = real exchange rate (increase = real depreciation); s_k^WS = Wine Spectator quality score.
  - Controls for firm marginal cost proxies z_i,t, year fixed effects, firm-destination fixed effects, product characteristics (grape, type, vintage), HS, and province fixed effects. Robust standard errors clustered at product level.
- Volume specification (gravity-consistent):
  - lnX_kij,t = α1 ln q_j,t + α2 s_k^WS + α3 (ln q_j,t × s_k^WS) + controls + fixed effects + error.
  - X = FOB export volume (liters); includes destination-year controls Z_j,t (real GDP, real effective exchange rate), firm controls and fixed effects.

### Main empirical findings — export prices and pass-through
- Baseline pass-through magnitude:
  - Following a ten percent change in the real exchange rate exporters change their export prices (in domestic currency) by 1.1 percent; implied pass-through = 89 percent.
- Heterogeneity by quality:
  - Interaction between ln(real exchange rate) and quality is positive and significant, consistent with Prediction 1: pricing-to-market increases with quality (pass-through decreases with quality).
  - Quantitative effect: a one standard deviation increase in quality from its mean level (a four point increase on the quality scale) increases pricing-to-market by five percent.
  - Pass-through by quality extremes:
    - Pass-through is complete (100 percent) for the wine with the lowest quality in the sample (score 55, elasticity = 0.065, insignificantly different from zero interpreted as full pass-through in context).
    - Pass-through drops to 86.5 percent for the highest quality wine (score 97; elasticity reported as positive and equal to 0.135).
- Robustness of price results:
  - Results remain when using alternative firm productivity proxy (number of export destinations), restricting to multi-product firms, and in a difference-in-difference specification with destination-year and product fixed effects (interaction remains positive though magnitude and significance fall).

### Main empirical findings — export volumes
- Aggregate volume responsiveness:
  - Export volumes respond positively to a real depreciation; estimated elasticity = 1.844 (Panel B, column (1)).
- Quality and volumes:
  - Coefficient on quality (within-firm effect) is negative and significant: within the same firm-destination, higher quality wines are exported in smaller quantities.
- Heterogeneity by quality:
  - Interaction between ln(real exchange rate) and quality is negative and significant, consistent with Prediction 2: the response of export volumes to a real depreciation decreases with quality.
  - Quantitative examples:
    - A one standard deviation increase in quality reduces the volume elasticity by one and a half percent from 1.833 to 1.808.
    - Elasticity for the highest quality wine = 1.75; for the lowest quality wine = 2.02.
- Robustness of volume results:
  - Negative interaction remains when using number of export destinations as firm performance proxy and when restricting to multi-product firms.
  - Difference-in-difference specification does not find significant heterogeneous volume responses driven by quality.

### Instrumentation and endogeneity
- Potential endogeneity: expert quality scores based on blind tastings but tasters know region/vintage which could bias scores.
- Instruments for quality: geography and weather-related factors — total rainfall and average temperatures during the growing season for each province, and altitude of each growing region.
- Empirical claim: main findings remain robust when quality is instrumented using these variables.
- IV diagnostics:
  - Kleibergen-Paap F statistic = 44 (rejects weak instruments; critical value 21 per Stock and Yogo, 2005).
  - IV results:
    - For unit values, coefficient on quality positive and significant but smaller than OLS (evidence of positive endogeneity bias).
    - For export volumes, instrumented effect of quality negative and significant, larger in magnitude than OLS.
    - Instrument first-stage patterns: February temperature positive; March temperature negative; October and December rainfall positive; January and February precipitation negative.

### Heterogeneity across destination countries (Predictions 3 and 4)
- Destination income split: World Bank 2011 GNI per capita threshold = $4,035 separates low and high income countries.
- Unit values (Panel A, Table 10):
  - Higher quality associated with higher prices.
  - Coefficients on real exchange rate for low and high income countries similar magnitude, but only high income coefficient significantly different from zero.
  - Pass-through:
    - Complete for low income destinations.
    - Decreases to approximately 90 percent for exports to high income countries.
  - Interaction of exchange rate with quality positive and significant for high income destinations only:
    - For high income countries, one standard deviation increase in quality increases exchange rate elasticity from 0.112 to 0.120 — a seven percent increase in pricing-to-market.
  - Robustness: holds when controlling for number of destinations, restricting to multi-product firms, and including destination-year and product fixed effects.
- Export volumes (Panel B, Table 10):
  - Real depreciation raises export volumes for both low and high income destinations.
  - Interaction between real exchange rate and quality significantly negative for high income countries only — response of export volumes to exchange rate changes smaller for higher quality wines.
  - Robust across specifications except when destination-year and product fixed effects included.

### Distribution costs and pricing-to-market
- Prediction: pricing-to-market increases with local distribution costs in importing economy; difference in pass-through between high and low quality wines should increase with distribution costs.
- Data and method:
  - Use Campa and Goldberg (2010) distribution costs for 21 countries and 29 industries (1995–2001).
  - Compute average distribution costs for each of 21 destination countries for “Food products and beverages”; denote dc_j.
  - Sample reduced by half due to limited country coverage; regressions include interaction between real exchange rate and (log) distribution costs.
- Findings:
  - Unit values: interaction between real exchange rate and distribution costs positive and significant at the ten percent level — pricing-to-market increases with local costs.
  - Interaction between real exchange rate and quality positive and significant — difference in pass-through between high and low quality wines increases with local costs.
  - Export volumes: interaction between exchange rate and local costs negative, significant at the 11 percent level.

### Aggregation bias and comparison with aggregate estimates
- Aggregation steps examined:
  - Aggregate at firm-destination-year level: unit values capture average price charged by each firm to each destination in each year; quality averaged by firm-destination-year s_ij,t,WS.
  - Aggregate across firms to destination-year: unit values average price of all firms to each destination in each year; quality denoted s_j,t,WS.
- Results (Table 11):
  - Benchmark product-level pass-through estimated at 89 percent (q_{j,t} elasticity = 0.112).
  - Firm-level average pass-through estimated at 69 percent (q_{j,t} elasticity ≈ 0.311).
  - Aggregate destination-year pass-through low at 47 percent (q_{j,t} elasticity ≈ 0.530).
- Conclusion: greater aggregation leads to lower estimated pass-through and makes identification of quality effects more difficult — evidence of aggregation bias.

### Robustness checks
- Measurement of quality:
  - Using log of quality s_k^WS: qualitatively unchanged; one standard deviation increase in quality from mean increases pricing-to-market by five percent.
  - Binned quality variable (1–6): qualitatively similar; magnitude larger.
  - Parker rankings s_k^P: results largely hold; coefficient on s_k^P larger than Wine Spectator.
  - Expand coverage by averaging quality by wine name and type to increase sample coverage from 58 percent to 85 percent of total FOB value; results qualitatively unaffected.
  - Export-volume regressions robust across quality measures.
- Endogeneity of quality:
  - Instruments: province average temperatures t_p,m (September–March), total rainfall r_p,m (September–March), and altitude Alt_p.
  - IV results replicate OLS findings over reduced samples; IV interaction results show pricing-to-market increases with quality (example: exchange rate elasticity increases from 0.105 to 0.131 following a one standard deviation increase in quality — 25 percent increase in pricing-to-market in one specification).
  - Instrumented interaction for volumes not always statistically significant.
- Asymmetries (appreciations vs depreciations):
  - Unit values: exchange rate elasticity at mean quality = 0.108 for real appreciations vs 0.092 for real depreciations (difference statistically significant at the one percent level).
  - One standard deviation increase in quality increases pricing-to-market by 8.3 percent following an appreciation and by 4.3 percent following a depreciation.
  - Export volumes: one standard deviation increase in quality reduces exchange rate elasticity more during depreciations (decrease of 2.1 percent) than appreciations (decrease of 0.5 percent).
- Sample period (post-2002):
  - Restricting sample to post-2002: one standard deviation increase in quality raises exchange rate elasticity from 0.099 to 0.106 — seven percent increase in pricing-to-market.
  - Elasticities slightly smaller than whole-sample estimates.
  - Example: exchange rate elasticity at mean quality equals 3.463 and drops to 3.440 following a one standard deviation increase in quality (both elasticities larger in some specifications).
- Effects of the 2008 Financial Crisis:
  - Pre-2008 sample: pass-through at mean quality estimated at 74 percent; drops to 73.4 percent after a one standard deviation increase in quality.
  - Interaction on export volumes negative but only significant at the ten percent level pre-2008.
- Intensive vs extensive margins:
  - Restricting to firms exporting in all years 2002–2009 (intensive margin only) yields robust results for prices and quantities.
- Currency of invoicing:
  - Datamyne (HS 2204) 2005–2008: 88 percent priced in US dollars; 7.6 percent euros; 3 percent Canadian dollars; 1.2 percent pound sterling.
  - Re-expressing unit values in US dollars per liter leaves results largely unaffected.
- Monthly frequency:
  - Re-estimating at monthly frequency: coefficient on quality decreases in magnitude but pricing-to-market remains significantly stronger for higher quality wines; export-volume results qualitatively similar.

### Price structure and industry relevance
- Empirical support that key model assumptions hold for wine:
  - Higher quality wines sold at higher prices.
  - Higher quality wines associated with higher marginal costs (e.g., more expensive inputs, oak barrels, drip irrigation, greater use of owned grapes).
  - Distribution costs and retailer margins increase with price/quality (UK non-EU example shows distribution costs and retailer margins rising across price categories).
- Price breakdown example (non-EU wine sold in UK retail outlets, per-bottle):
  - Retail prices: £5.76; £7.19; £8.83; £10.09.
  - VAT (20%): £0.96; £1.20; £1.47; £1.68.
  - Retail margin: £1.92; £2.40; £2.94; £3.36.
  - Duty: £1.90 (constant).
  - Distributor margin: £0.11; £0.21; £0.40; £0.51.
  - Common Customs Tariff: £0.11 (constant).
  - Transport: £0.13 (constant).
  - Winemaker share: £0.63; £1.25; £1.88; £2.40.

### Summary of substantive findings and implications
- Quality matters for exchange rate pass-through:
  - Pricing-to-market (elasticity of export prices to real depreciation) increases with product quality; pass-through decreases with quality.
  - Export volumes increase after a real depreciation, but the increase is smaller for higher quality products.
  - Heterogeneity by destination income: quality-driven effects stronger for high income destination countries.
- Policy and empirical implications:
  - Aggregate estimates of exchange rate pass-through may understate firm-product level responsiveness due to aggregation bias.
  - Understanding quality heterogeneity is important for monetary policy transmission and exchange rate policy implications on trade balances because quality influences how exporters adjust prices and volumes to currency movements.

### Selected key statistics and regression estimates
- Baseline pass-through:
  - Following a ten percent real depreciation exporters increase export prices by 1.1 percent → pass-through = 89 percent.
- Baseline elasticities (Table 8):
  - Panel A (ln UV): coefficient on ln q_{j,t} = 0.112 (t = 2.947).
  - Panel A: s^k_{WS} = 0.032 (t = 12.178); ln q_{j,t} × s^k_{WS} = −0.002 (t = 3.042).
  - Panel B (ln X): coefficient on ln q_{j,t} = 1.844 (t = 3.550).
  - Panel B: s^k_{WS} = −0.046 (t = −7.423); ln q_{j,t} × s^k_{WS} = −0.006 (t = −3.742).
- Aggregation bias (Table 11):
  - Product-level q_{j,t} elasticity = 0.112 → Pass-through = 89%.
  - Firm-level q_{j,t} elasticity ≈ 0.311 → Pass-through = 69%.
  - Aggregate q_{j,t} elasticity ≈ 0.530 → Pass-through = 47%.
- Data coverage (trade panel totals):
  - Total observations: 41,576.
  - FOB exports (USD): 725,789,235.
  - Firms: 2,096.
  - Wines: 6,720.
- Top export destinations (2002–2009 shares of FOB exports):
  - United States: 30.8 percent.
  - Netherlands: 10.3 percent.
  - United Kingdom: 9.3 percent.
  - Brazil: 7.2 percent.
  - Canada: 6.4 percent.
  - Denmark: 6.0 percent.

*Source: _wp1442 - 15. Robustness for Export Volumes*

### References .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .

### References

### Tables
- 1.Harmonized System (HS) Classification Codes .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    46
- 2.Summary Statistics on Trade Data by Year  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    46
- 3.Summary Statistics   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    47
- 4.Top Export Destinations 2002-2009    .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    47
- 5.Experts Ratings   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    48
- 6.Snapshot of the Data    .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    48
- 7.Price Breakdown for Non-EU Wine sold in Retail Outlets in the UK   .  .  .  .  .  .  .  .  .    48
- 8.Baseline Results  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    49
- 9.Local Distribution Costs    .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    50
- 10.    Heterogeneity across Destination Countries    .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    51
- 11.    Aggregation Bias   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    52
- 12.    Unit Values: Robustness on Quality .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    53
- 13.    Export Volumes: Robustness on Quality   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    54
- 14.    Robustness for Unit Values   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    55

*Source: _wp1442 - References*

### 15.    Robustness for Export Volumes  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .

### 15.    Robustness for Export Volumes

### Introduction and motivation
- Empirical background: pass-through of exchange rate changes to domestic/import prices is typically incomplete; incomplete pass-through implies deviations from the Law of One Price and has implications for monetary and exchange rate policy.
- Research gap: scarce evidence on the role of product-level quality in explaining heterogeneous exchange rate pass-through.
- Strategy: focus on the wine industry where “quality” is observable via expert ratings; combine Argentinean firm-level destination-specific export values and volumes with Wine Spectator and Robert Parker ratings.

### Data and sample
- Customs trade data (Argentina) cover export flows between 2002 and 2009 with exporter name, destination country, date, 12-digit HS code, FOB value (US dollars), and volume (liters); product detail includes wine name, grape, type, vintage year.
- Final cleaned sample: 21,647 different products/wines overall; 6,720 wines matched with quality rankings (31 percent of wines, representing 58 percent of total FOB value exported between 2002 and 2009).
- Firms and scope: 6,720 wines exported by 209 wine producers; exporters exported an average of 139 different wines (range 1 to 510); firms exported to an average of 40 different destinations (range 1 to 88).
- Quality measures: Wine Spectator scores on a (50,100) scale (mean ranking 85; min 55; max 97). Parker rankings also on a (50,100) scale (sample: 3,969 wines, mean 87, min 72, max 98). Overlap: 2,433 wines have both rankings; correlation between the two rankings is 0.53.
- Macroeconomic data: GDP from Penn World Tables; CPI and nominal exchange rates from IFS; real exchange rate defined as ratio of CPIs times average yearly nominal exchange rate (an increase captures a real depreciation). Argentina experienced a peso depreciation of up to 75 percent in 2002 after the 2001 financial crisis; further depreciation in 2008.

### Theoretical framework and testable predictions
- Model setup: extends Corsetti and Dedola (2005) to multi-product firms heterogeneous in product quality; local additive distribution costs paid in importing-country currency imply distribution costs are unaffected by exchange rate changes.
- Key model implications:
  - Demand elasticity perceived by firms falls with a real depreciation and with quality.
  - Higher quality goods have higher marginal costs and higher markups.
- Predictions (as derived in the model):
  - Prediction 1: the elasticity of export price p_j(φ) to a change in the real exchange rate q_j (pricing-to-market) increases with product quality s(φ).
  - Prediction 2: the elasticity of export volumes x_j(φ) to a change in q_j decreases with product quality s(φ).
  - Prediction 3 & 4 (non-homothetic preferences): the effects in Predictions 1 and 2 are stronger for high income destination countries (preference for quality increases with per capita income).

### Empirical framework
- Price (unit value) specification (reduced-form):
  - lnUV_kij,t = β1 ln q_j,t + β2 s_k^WS + β3 (ln q_j,t × s_k^WS) + controls + fixed effects + error.
  - UV = export unit value in pesos per liter (proxy for export price); q_j,t = real exchange rate (increase = real depreciation); s_k^WS = Wine Spectator quality score.
  - Controls for firm marginal cost proxies z_i,t (firm total export volume and number of destinations), year fixed effects, firm-destination fixed effects, product characteristics (grape, type, vintage), HS, and province fixed effects. Robust standard errors clustered at product level.
- Volume specification (gravity-consistent):
  - lnX_kij,t = α1 ln q_j,t + α2 s_k^WS + α3 (ln q_j,t × s_k^WS) + controls + fixed effects + error.
  - X = FOB export volume (liters); includes destination-year controls Z_j,t (real GDP, real effective exchange rate), firm controls and fixed effects.

### Main empirical findings — export prices and pass-through
- Baseline pass-through magnitude:
  - In the baseline regression, following a ten percent change in the real exchange rate exporters change their export prices (in domestic currency) by 1.1 percent; implied pass-through = 89 percent.
- Heterogeneity by quality:
  - The interaction coefficient between ln(real exchange rate) and quality is positive and significant, consistent with Prediction 1: pricing-to-market increases with quality (i.e., pass-through decreases with quality).
  - Quantitative effect: a one standard deviation increase in quality from its mean level (a four point increase on the quality scale) increases pricing-to-market by five percent.
  - Pass-through by quality extremes: pass-through is complete (100 percent) for the wine with the lowest quality in the sample (score 55, elasticity = 0.065, insignificantly different from zero interpreted as full pass-through in context), but drops to 86.5 percent for the highest quality wine (score 97; elasticity reported as positive and equal to 0.135).
- Robustness of price results:
  - Results remain when using alternative firm productivity proxy (number of export destinations), restricting to multi-product firms, and in a difference-in-difference specification with destination-year and product fixed effects (interaction remains positive though magnitude and significance fall).

### Main empirical findings — export volumes
- Aggregate volume responsiveness:
  - Export volumes respond positively to a real depreciation; estimated elasticity (column (1) of Panel B) = 1.844.
  - The coefficient on quality (within-firm effect) is negative and significant: within the same firm-destination, higher quality wines are exported in smaller quantities.
- Heterogeneity by quality:
  - The interaction between ln(real exchange rate) and quality is negative and significant, consistent with Prediction 2: the response of export volumes to a real depreciation decreases with quality.
- Robustness of volume results:
  - The negative interaction remains when using the number of export destinations as firm performance proxy and when restricting to multi-product firms. The difference-in-difference specification does not find significant heterogeneous volume responses driven by quality.

### Instrumentation and endogeneity
- Potential endogeneity: expert quality scores are based on blind tastings but tasters know region/vintage which could bias scores.
- Instruments used for quality (to address endogeneity): geography and weather-related factors — total rainfall and average temperatures during the growing season for each province where grapes are grown, and altitude of each growing region.
- Empirical claim: main findings remain robust when quality is instrumented using these variables.

### Aggregation bias and comparison with aggregate estimates
- Aggregation bias: aggregating data at firm or country levels reduces estimated pass-through — “the more aggregated the data, the lower is estimated pass-through,” suggesting aggregate pass-through estimates suffer from aggregation bias.
- Comparison with aggregate/industry estimates: Campa and Goldberg (2005) report average pass-through of 46 percent in the short run and 64 percent in the long run for a sample of OECD countries; the firm-level pass-through of 89 percent found here contrasts with lower aggregate estimates.

### Relevance of the wine industry to the model assumptions
- Empirical support that key model assumptions hold for wine:
  - Higher quality wines are sold at higher prices (consistent with equation (7)).
  - Higher quality wines are associated with higher marginal costs — examples: more expensive inputs, oak barrels, drip irrigation, greater use of owned grapes for high-quality wines.
  - Distribution costs and retailer margins increase with price/quality (UK non-EU example: distribution costs and retailer margins both rise across price categories).
  - Conclusion: wine industry features closely satisfy the model assumptions (higher quality → higher marginal costs, higher distribution costs, higher markups).

### Summary of substantive findings and implications
- Quality matters for exchange rate pass-through:
  - Pricing-to-market (elasticity of export prices to real depreciation) increases with product quality; pass-through decreases with quality.
  - Export volumes increase after a real depreciation, but the increase is smaller for higher quality products.
  - Heterogeneity by destination income: model predicts (and empirical analysis explores) stronger quality-driven effects for high income destination countries due to stronger preferences for quality.
- Policy and empirical implications:
  - Aggregate estimates of exchange rate pass-through may understate the responsiveness of prices at the firm-product level due to aggregation bias.
  - Understanding quality heterogeneity is important for monetary policy transmission and for exchange rate policy implications on trade balances because quality influences how exporters adjust prices and volumes to currency movements.

*Source: _wp1442 - 15. Robustness for Export Volumes (extracted from the supplied PDF chapter content).*

### 86.5 percent. In column (2) of Panel B, a one standard deviation increase in quality reduces

### _wp1442 - 86.5 percent. In column (2) of Panel B, a one standard deviation increase in quality reduces

### Effects of quality on price and volume elasticities
- A one standard deviation increase in quality reduces the volume elasticity by one and a half percent from 1.833 to 1.808.
- The elasticity for the highest quality wine is equal to 1.75 and increases to 2.02 for the lowest quality wine.
- Effects of quality on price and volume elasticities remain very similar across all specifications reported in Table 8.

### Distribution costs and pricing-to-market
- Model prediction (Corsetti and Dedola, 2005 extension): pricing-to-market increases with local distribution costs in the importing economy; hence difference in pass-through between high and low quality wines should increase with distribution costs.
- Data and method:
  - Use Campa and Goldberg (2010) distribution costs for 21 countries and 29 industries between 1995 and 2001.
  - Compute average distribution costs over time for each of the 21 destination countries and for the “Food products and beverages” industry; denote destination-specific measure as dc_j.
  - Sample size reduced by half due to limited country coverage.
  - Estimate regressions (11) and (12) including interaction between real exchange rate and (log) distribution costs.
- Findings (Table 9):
  - Unit values: interaction between real exchange rate and distribution costs is positive and significant at the ten percent level — pricing-to-market increases with local costs.
  - Column (2): interaction between real exchange rate and quality is positive and significant — difference in pass-through between high and low quality wines increases with local costs.
  - Export volumes (columns (3) and (4)): interaction between exchange rate and local costs is negative, significant at the 11 percent level.

### Heterogeneity across destination countries (Predictions 3 and 4)
- Destination split: World Bank 2011 GNI per capita threshold of $4,035 separates low and high income countries.
- Method: estimate equations (11) and (12) for unit values and export volumes; interact real exchange rate and its interaction with quality with High and Low destination dummies.
- Unit values (Panel A, Table 10):
  - Higher quality associated with higher prices.
  - Coefficients on real exchange rate for low and high income countries are of similar magnitude, but only high income coefficient is significantly different from zero.
  - Pass-through:
    - Estimated as complete for low income destinations.
    - Decreases to approximately 90 percent for exports to high income countries — indicates price discrimination across destinations and failure of the Law of One Price.
  - Column (2): interaction of exchange rate with quality is positive and significant for high income destinations only.
    - For high income countries: one standard deviation increase in quality increases the exchange rate elasticity from 0.112 to 0.120, i.e., a seven percent increase in pricing-to-market.
  - Robustness: finding holds when controlling for number of destinations (column 3), restricting to multi-product firms (column 4), and including destination-year and product fixed effects (column 5).
- Export volumes (Panel B, Table 10):
  - Real depreciation raises export volumes for both low and high income destinations.
  - Column (2): interaction between real exchange rate and quality is significantly different from zero for high income countries only; negative sign indicates response of export volumes to exchange rate changes is smaller for higher quality wines.
  - Robust across specifications except when destination-year and product fixed effects included (column 5).

### Aggregation bias
- Motivation: firm-level pass-through estimates are larger than many aggregate/industry-level estimates.
- Aggregation steps:
  - Aggregate at firm-destination-year level (firm-level single-product): unit values capture average price charged by each firm to each destination in each year; quality is average quality of wines exported by each firm to each destination in each year, denoted s_i j,t,WS.
  - Further aggregate across firms to destination-year: unit values are average price of all firms to each destination in each year; quality denoted s_j,t,WS.
- Results (Table 11):
  - Columns (1)-(2): benchmark specifications for unit values — average pass-through estimated at 89 percent whether or not interacting real exchange rate with quality.
  - Columns (3)-(4): firm-level sample — higher quality associated with higher prices; interaction insignificant (likely due to reduced variation in quality). Average pass-through estimated at 69 percent (20 percentage points lower than 89 percent).
  - Columns (5)-(6): aggregate sample — higher quality still associated with higher unit values but significance reduced; interaction insignificant; average pass-through low at 47 percent.
- Conclusion: part of low pass-through estimated using aggregate data appears due to aggregation bias — more aggregation yields lower estimated pass-through and makes identification of quality effects more difficult.
- Additional notes:
  - Majority of firms export in a single HS code, motivating firm-level aggregation rather than HS-level.
  - Standard error and fixed-effect details described for each aggregation level.

### Robustness checks
A. Measurement of quality
- Alternatives tested (Table 12):
  - Use log of quality s_k^WS: qualitatively unchanged; one standard deviation increase in quality from mean increases pricing-to-market by five percent.
  - Create 1–6 binned quality variable ˜s_k^WS (Wine Spectator bins): results qualitatively similar; magnitude of quality coefficient larger.
  - Use Parker rankings s_k^P: results largely hold; coefficient on s_k^P larger than Wine Spectator rankings.
  - Address missing Wine Spectator observations by averaging quality by wine name and type to increase sample coverage from 58 percent to 85 percent of total FOB value (denoted s_K^WS and ˜s_K^WS); results qualitatively unaffected.
- Table 13: using export volumes as dependent variable yields robust results across these measures.

B. Endogeneity of quality
- Concern: Wine Spectator and Parker rankings are produced from blind tastings where price is not taken into account, but tasters know general type and vintage — potential endogeneity.
- Instruments: geographic and weather-related factors — average temperature t_p,m and total rainfall r_p,m for growing months September–March, and altitude Alt_p of each province.
  - Data sources: National Climatic Data Center; gaps filled via online sources; some missing observations reduce sample.
  - Province-specific weather stations listed for temperature/rainfall measurement.
- IV results (Tables 12 and 13):
  - Replicate benchmark OLS over reduced sample (column (7)): main findings hold.
  - IV for unit values (column (8), Table 12): coefficient on quality positive and significant but smaller than OLS — indicates positive endogeneity bias (tasters assign higher scores to more expensive wines).
  - IV for export volumes (column (8), Table 13): instrumented effect of quality on export volumes negative and significant, larger in magnitude than OLS.
  - Kleibergen-Paap F statistic equal to 44 for both prices and quantities regressions (critical value 21, Stock and Yogo, 2005) — rejects weak instruments.
  - First-stage patterns: February temperature positive; March temperature negative (counterintuitive); October and December rainfall positive; January and February precipitation negative — broadly consistent with literature expectations.
  - Instrument set expanded to include interactions of instruments with exchange rate when instrumenting quality interacted with exchange rate.
  - Column (9), Table 12 (unit values): exchange rate elasticity increases from 0.105 to 0.131 following a one standard deviation increase in quality from its mean level — corresponds to a 25 percent increase in pricing-to-market.
  - Column (9), Table 13 (export volumes): coefficient on the interaction term not statistically significant.

C. Asymmetries (appreciations vs depreciations)
- Method: interact real exchange rate and its interaction with quality with dummies for appreciations (App) and depreciations (Dep).
- Unit values (Table 14, column (1)):
  - Exchange rate elasticity at mean quality: 0.108 for real appreciations vs 0.092 for real depreciations — difference statistically significant at the one percent level.
  - One standard deviation increase in quality from mean increases pricing-to-market by 8.3 percent following a real appreciation and by 4.3 percent following a real depreciation.
- Export volumes (Table 15, column (1)):
  - One standard deviation increase in quality reduces the exchange rate elasticity more during depreciations (decrease of 2.1 percent) than appreciations (decrease of 0.5 percent).
- Interpretation: asymmetry consistent with firms reducing domestic currency prices more when peso appreciates to maintain market shares; consistent with Marston (1990) evidence.

D. Sample period (post-2002)
- Background: fixed exchange rate regime abandoned in 2001; large peso depreciation in 2002.
- Results (Table 14, column (2)):
  - Restricting sample to post-2002: one standard deviation increase in quality from mean raises exchange rate elasticity from 0.099 to 0.106 — seven percent increase in pricing-to-market.
  - Elasticities slightly smaller than whole-sample estimates — suggests pricing-to-market stronger when exchange rates are more variable (consistent with Martín and Rodríguez, 2004).
- Export volumes (Table 15, column (2)):
  - Exchange rate elasticity at mean quality equals 3.463 and drops to [text truncated in source at this point].

*Source: IMF working paper content provided in the input PDF excerpt.*

### 3.440 following a one standard deviation increase in quality, where both elasticities are larger

### _wp1442 - 3.440 following a one standard deviation increase in quality, where both elasticities are larger

### Effects of the 2008 Financial Crisis
- Peso depreciated with respect to the US dollar after 2008; consumers may have substituted toward lower quality imports ("flight from quality effect").
- Pre-2008 sample results (column (3) of Table 14):
  - Elasticity of export prices to exchange rate changes remains significantly larger for higher quality wines.
  - Pass-through is estimated at 74 percent at the mean level of quality.
  - Pass-through drops to 73.4 percent after a one standard deviation increase in quality.
- Interaction between real exchange rate and quality on export volumes in the pre-2008 sample is negative but only significant at the ten percent level.
- Interpretation: before the crisis Argentinean exporters priced-to-market more; increased financial constraints during/post-crisis may have reduced pricing-to-market (Strasser, 2013).

### Extensive Margin vs Intensive Margin
- Depreciation affects average prices via intensive margin (existing exporters) and can induce entry via extensive margin; entrants are generally less productive and charge higher prices, reducing observed pass-through.
- Robustness check restricting to firms that export in all years 2002–2009 (captures intensive margin only) — results for prices and quantities (column (4) of Tables 14 and 15) remain robust to exclusion of the extensive margin.

### Role of the US Dollar and Destination Countries
- After the 2002 devaluation, peso was allowed to fluctuate within a crawling band "narrower than or equal to +/-2 percent" with respect to the US dollar.
- Variations in the real exchange rate between peso and US dollar may largely reflect domestic price movements.
- Excluding the US and currencies pegged to the US dollar (list of countries per Reinhart and Rogoff, 2004) — results reported in columns (5) and (6) of Tables 14 and 15 — remain largely robust.

### Argentinean Consumer Price Index (CPI) Concerns
- Since 2007, official Argentinean CPI credibility questioned; IMF issued a declaration of censure.
- Alternative sources show considerably higher inflation since 2007; official real exchange rate for 2008 and 2009 may be unreliable.
- Cavallo (2013) constructs an aggregate CPI using supermarket data showing inflation about three times larger than official estimate.
- Using Cavallo CPI for 2008 and 2009 to update official CPI and constructing new real exchange rate q̃_j,t for equations (11) and (12): results reported in column (7) of Tables 14 and 15 remain very similar to those with official CPI.

### Monthly Frequency Robustness
- Data available at transaction (date of declaration) level; regressions re-estimated at monthly frequency with unit values UV_kijm, export volumes X_kijm, and real exchange rate q_j,m, month fixed effects.
- Due to data limits, GDPs and real effective exchange rates in exports regression remain annual; average firm size defined yearly.
- Column (8) of Table 14 (unit values): coefficient on quality decreases in magnitude but pricing-to-market remains significantly stronger for higher quality wines.
- Column (8) of Table 15 (export volumes): qualitatively similar to annual frequency; coefficient on quality smaller; coefficient on real exchange rate reduced but export volumes still increase in response to real depreciation and by less for higher quality wines.

### Currency of Invoicing
- No firm-level invoicing currency in main data; Datamyne (sector HS 2204) provides invoicing currency for 2005–2008:
  - 88 percent priced in US dollars.
  - 7.6 percent priced in euros.
  - 3 percent priced in Canadian dollars.
  - 1.2 percent priced in pound sterling.
  - Very few priced in Japanese yen, Swiss francs, Uruguayan pesos, Australian dollars, or Danish krones.
- Re-expressing unit values in US dollars per liter (column (9) of Table 14) leaves results largely unaffected.

### Model Contributions and Mechanisms
- Theoretical model builds on Berman and others (2012) and Chatterjee and others (2013), extends Corsetti and Dedola (2005) by allowing firms to export multiple products with heterogeneous quality.
- With additive local distribution costs paid in importer's currency, perceived demand elasticity falls with real depreciation and with quality.
- Prediction: exporters increase prices more and export volumes less in response to real depreciation for higher quality goods.
- Heterogeneous response predicted to be stronger for exports to high income destination countries when richer countries prefer higher quality goods.

### Empirical Strategy and Data
- Combined unique firm-level destination-specific export values and volumes for highly disaggregated wine products, 2002–2009, with two expert wine ratings (Wine Spectator and Robert Parker) to measure quality.
- Baseline sample includes 6,720 different wine products exported by 209 wine producers over the period.

### Key Empirical Findings
- Pass-through: "following a ten percent real depreciation exporters increase their export prices by 1.1 percent so pass-through is 89 percent."
- Higher quality associated with higher prices.
- Response of export prices to real exchange rate changes increases with quality; quality explains large part of heterogeneous pass-through.
- Export volumes increase following a real depreciation, but the increase is smaller for higher quality wines.
- Heterogeneous response of prices and quantities to real exchange rate fluctuations present only when firms export to high income destination countries.
- Results robust to different measures of quality, samples, specifications, potential endogeneity of quality, and accounting for aggregation bias.

*Source: _wp1442 - 3.440 following a one standard deviation increase in quality, where both elasticities are larger*

### REFERENCES

### _wp1442 - REFERENCES

### Major cited literature (selected topics)
- Importers, exporters, and exchange rate pass-through; pricing-to-market; firm heterogeneity; product quality and exports; distribution costs and real exchange rate dynamics.
- Representative citations include Amiti, Itskhoki and Konings (2012); Atkeson and Burstein (2008); Auer and Chaney (2009); Berman, Martin and Mayer (2012); Burstein and Gopinath (2013); Campa and Goldberg (2005, 2010); Gopinath and Itskhoki (2010); Hallak (2006); Hummels and Klenow (2005); Manova and Zhang (2012a,b); Melitz and Ottaviano (2008); Nakamura and Steinsson (2012); Reinhart and Rogoff (2004); Verhoogen (2008); Yu (2013).
- Empirical and theoretical approaches are both represented, including working papers, CEPR and NBER papers, Journal articles (e.g., American Economic Review, Quarterly Journal of Economics, Review of Economic Studies), and mimeos.

### Data coverage and descriptive statistics
- Trade panel: Observations by year
  - 2002: 2,067 observations; FOB exports (USD) 36,504,644; Firms 597; Wines 94
  - 2003: 3,056 observations; FOB exports (USD) 50,664,899; Firms 739; Wines 33
  - 2004: 3,923 observations; FOB exports (USD) 69,640,144; Firms 1,071; Wines 1,171
  - 2005: 5,330 observations; FOB exports (USD) 91,261,787; Firms 1,201; Wines 1,517
  - 2006: 6,793 observations; FOB exports (USD) 112,540,681; Firms 1,501; Wines 1,731
  - 2007: 7,407 observations; FOB exports (USD) 131,147,970; Firms 1,481; Wines 1,860
  - 2008: 6,865 observations; FOB exports (USD) 125,851,505; Firms 1,481; Wines 1,804
  - 2009: 6,135 observations; FOB exports (USD) 108,177,602; Firms 1,511; Wines 1,833
  - Total: 41,576 observations; FOB exports (USD) 725,789,235; Firms 2,096; Wines 6,720
  - Note: "Authors’ own calculations."
- Table 3 summary statistics (N = 41,576 unless noted)
  - Unit values (USD/liter): Min 0.0238; Max 15.33; Mean 3.6; Median 6.7
  - Number of wines exported: Min 1; Max 5; Mean 3; Median 1 (table shows "151013912030" — interpreted as the row entries but source text preserved)
  - Number of destinations: Min 1; Max 88; Mean 40; Median 37 (table shows "188403723")
  - Wine Spectator: Min 55; Max 99; Mean 85; Median 85.3
  - Parker: Min 72; Max 98; Mean 87; Median 72.3
  - Instrument variables (province-level climate and altitude): Temperature and rainfall monthly statistics (September–March) and Altitude (meters): Min 19; Max 1,238; Mean 716; Median 705; Std dev 134 (row preserved as in source).
  - Notes: "Authors’ own calculations."

### Top export destinations (2002–2009)
- United States: 30.8 percent of FOB exports
- Netherlands: 10.3 percent
- United Kingdom: 9.3 percent
- Brazil: 7.2 percent
- Canada: 6.4 percent
- Denmark: 6.0 percent
- Finland: 3.2 percent
- Sweden: 3.2 percent
- Switzerland: 2.8 percent
- Germany: 2.3 percent
- France: 1.8 percent
- Notes: "Authors’ own calculations."

### Expert rating scales
- Wine Spectator (50,100)
  - 95-100 Great
  - 90-94 Outstanding
  - 85-89 Very good
  - 80-84 Good
  - 75-79 Mediocre
  - 50-74 Not recommended
- Parker (50,100)
  - 96-100 Extraordinary
  - 90-95 Outstanding
  - 80-89 Above average/very good
  - 70-79 Average
  - 60-69 Below average
  - 50-59 Unacceptable

### Price structure example (Table 7): Non-EU wine sold in UK retail outlets (per-bottle breakdown)
- Retail prices considered: £5.76; £7.19; £8.83; £10.09
- VAT (20%): £0.96; £1.20; £1.47; £1.68
- Retail margin: £1.92; £2.40; £2.94; £3.36
- Duty: £1.90 (constant across prices)
- Distributor margin: £0.11; £0.21; £0.40; £0.51
- Common Customs Tariff: £0.11 (constant)
- Transport: £0.13 (constant)
- Winemaker share: £0.63; £1.25; £1.88; £2.40
- Source: Joseph (2012).

### Key empirical estimates (selected regression results; unit values in pesos per liter; export volumes in liters)
- Table 8. Baseline results (Panel A: dependent variable ln UV; Panel B: dependent variable ln X)
  - Panel A: coefficient on ln q_{j,t} = 0.112 (t = 2.947) in column (1)
  - Panel A: s^k_{WS} = 0.032 (t = 12.178) and ln q_{j,t} × s^k_{WS} = −0.002 (t = 3.042)
  - Panel B: coefficient on ln q_{j,t} = 1.844 (t = 3.550) in column (1)
  - Panel B: s^k_{WS} = −0.046 (t = −7.423) and ln q_{j,t} × s^k_{WS} = −0.006 (t = −3.742)
  - Quantitative effects reported (Panel A): Mean(s) –0.115 (t = 3.02)a ; Mean(s)+sd(s) –0.121 (t = 3.19)a
  - Quantitative effects reported (Panel B): Mean(s) –1.833 (t = 3.54)a ; Mean(s)+sd(s) –1.808 (t = 3.48)a
  - Sample sizes: N = 41,576 (full sample in columns shown)
  - Notes: Year, province, firm-destination, grape, type, vintage year, and HS fixed effects included in (1)-(4); year-destination and product fixed effects included in (5). Robust standard errors clustered at product level. a, b, c indicate significance at 1, 5, and 10 percent levels, respectively.

- Table 11. Aggregation bias (q_{j,t} elasticity and implied pass-through)
  - Column (1) q_{j,t} elasticity = 0.112; Pass-through = 89%
  - Column (2) q_{j,t} elasticity = 0.115; Pass-through = 89%
  - Column (3) q_{j,t} elasticity = 0.311; Pass-through = 69%
  - Column (4) q_{j,t} elasticity = 0.306; Pass-through = 69%
  - Column (5) q_{j,t} elasticity = 0.530; Pass-through = 47%
  - Column (6) q_{j,t} elasticity = 0.530; Pass-through = 47%
  - Samples: Products (N = 41,576, 41,576), Firms (N = 6,505, 6,505), Aggregate (N = 510, 510)

- Tables 12–15. Robustness and IV evidence (selected diagnostics and sample sizes)
  - Table 12 (Unit values, IV sample): Kleibergen-Paap F = 44.15 and 22.35 reported for IV specifications
  - Table 12, sample sizes: N up to 67,585 in extended samples; IV sample N = 39,146 in some specifications
  - Table 13 (Export volumes, IV sample): Kleibergen-Paap F = 44.20 and 22.33 reported; IV sample N = 39,146
  - Many robustness checks across subsamples: Full; Post-2002; Pre-2008; Intensive; No US; No USD peg; Monthly.
  - Reported quantitative effects for ln X (mean(s)) range across specifications, e.g., 1.832 (t = 3.53)a, 3.106 (t = 4.62)a, 1.827 (t = 3.20)a, etc., depending on specification.

### Tables and figures included in the content unit
- Reference list (multiple pages of citations).
- Appendix tables:
  - Table 1. Harmonized System (HS) Classification Codes (6-digit/12-digit descriptions for wine categories)
  - Table 2. Summary Statistics on Trade Data by Year
  - Table 3. Summary Statistics
  - Table 4. Top Export Destinations 2002-2009
  - Table 5. Experts Ratings (Wine Spectator; Parker)
  - Table 6. Snapshot of the Data (example firm-year-destination-wine rows with unit values in USD/liter)
  - Table 7. Price Breakdown for Non-EU Wine sold in Retail Outlets in the UK (detailed per-bottle components)
  - Tables 8–15. Regression results, heterogeneity, aggregation bias, robustness checks, and diagnostics (detailed coefficient estimates, t-statistics, sample sizes, fixed effects, clustering conventions)
- Figures referenced: 
  - Figure 1. Argentina’s Total Wine Exports (million US dollars)
  - Figure 2. Wine Spectator versus Parker rankings
  - Figure 3. Argentinean peso per US dollar, January 2002 to December 2009

*Source: _wp1442 - REFERENCES (IMF working paper appendix and reference list).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1442.pdf_